Las notas de release están escritas en inglés. La página está traducida.
v2.54.0
NetPulse 2.54.0
This release is about what the screen was not telling you. Four features make the panel explain more, and a batch of fixes make it stop lying in the places where it was.
Icons that name the device
Searching the icon picker by the word Fridge and the row taking the new icon
Sixty smart-home icons in ten categories, searchable by their name in your language. A fridge stops looking like a generic box, and the search finds it by typing what it is. The catalog came from @Bann-DEV in #2107, and #2113 is the test that keeps it honest when someone adds a tile.
What each device asks for
DNS activity for one device with the 1 hour and 24 hour windows
Per-device DNS activity from the AdGuard query log: the domains a client resolves, how often, and which ones were blocked, in a one hour or one day window. It sits in the device detail, next to the client that asked. #2081 built it, and #2119 made it show something, because it was reading with a lower case MAC while the log was stored in upper case.
Quiet is not offline
A neighbour entry in STALE coming back to REACHABLE and the host staying online
A wired host with a static address that goes quiet used to drop off the map. Now the unit pings its stale neighbour and re-reads the table, so the address and the MAC are confirmed and an idle host stays online. #2108 added the recheck, #2109 extended it to IPv6 over NDP, and #2110 sweeps the subnets a unit routes to find hosts that never left an entry at all. The report behind it is #2063.
Chinese and Russian
The same panel in English, Spanish, Chinese and Russian, with dates following each language
The whole interface in Chinese and in Russian, catalogs complete and reviewed. Dates and numbers follow each language too, not only the labels. Chinese came in #2070, from @Bann-DEV, and Russian in #2083, from @sers88.
Also in this release
- What's new page. The app now explains each release with animated cards, in all four languages. #2116 added it, and the dot on the version stamp stays until you open it.
- Traffic per client. The device detail shows down, up and total over 1 hour, 24 hours, 30 days and 12 months, with the series for the same window. @tombacchetta asked for it in #2076 and the panel answers with the per-client history the server already had.
- Wi-Fi clients with no IP are marked as such, and leave one info entry in the activity log after two minutes instead of looking broken. #2111 closed the request @crowedavid opened in #1984.
- Docker. An optional server image with the UI and the agents inside, published to GHCR on every tag for amd64 and arm64. @sers88 wrote it in #2101 and the pipeline in #2105.
- A smaller agent. Built without
encoding/json/v2: 7.8 MB down to 7.4 MB on amd64, which is 5.8% less. #2096 measured it before and after. - The client table rebalances its columns and uses a single line in the 2xl tier, from #2106.
- A quieter panel. The SNMP community field says it is read-only (#2080), the unit column is translated (#2098), the splash background follows the theme (#2099) and the landing page counts demo clicks (#2093).
Fixed
- The per-device DNS panel showed nothing until #2119 fixed the case mismatch between what ingest stored and what the request asked for.
- Responses from handlers that wait on a unit died at the 15 second write timeout, so a reinstall that worked still showed a red "server unreachable" line. #2112 gave each handler its own ceiling, and it came from the install report in #1894.
- The rack offers cables between fleet units again, from the FDB of the designated unit. The interface and the implementation had drifted apart, and #2120 put them back together.
- MACs are normalised with one parser at the boundaries in #2121, which was losing GL clients with hyphenated addresses and breaking the
?mac=filters. #2134 closes the same hole on ingest, for any pusher that sends lower case or hyphenated addresses. - Discovery no longer marks an authorised host as "Key missing" when its board has no
modelfield, the case @xvlvx0 described in #2122. - A chain of nested device hubs draws every level, each device once, and its cable is emitted once. #2123 and #2124 are the two halves of the same bug, reported by @spskeldon.
- A bridge unit only names the peer on a port when its FDB makes it unambiguous, which is what @crowedavid asked for in #1976 after a port insisted it was connected to the wrong switch.
- owut stops aborting because
-ris repeated when the exclusion list is already in UCI, tracked in #2102. - A polled switch is no longer drawn twice in the topology, from #2104.
- BATMAN-adv LOOP DETECT frames stop showing up as clients and stop holding topology nodes. @tam481 explained them in #1818 and @crowedavid confirmed they were mesh probes, not hosts.
- Data tables hidden for screen readers no longer stretch the page scroll area, from #2103.
Thanks
- @Bann-DEV for the icon catalog in #2107 and the Chinese translation in #2070, and for #2113, the test that keeps the catalog honest.
- @sers88 for the Docker image in #2101 and its pipeline in #2105, which turned a request into an install path, and for the Russian catalog in #2083.
- @tam481 for the report in #1818 that explained a client list with more than a thousand phantoms, and @crowedavid for recognising what those frames were.
- @crowedavid again, for #1976, a port that named the wrong switch, and for the install report in #1894 that led to #2112.
- @spskeldon for #2123 and #2124, the two halves of the nested hub bug.
- @tombacchetta for #2076, asking for traffic per client and getting a panel that answers it.
- @xvlvx0 for #2122, where an authorised host was labelled as missing its key.
- @mgrupenhoff for #2119, #2120 and #2121, three fixes at the MAC boundaries that were losing data quietly.
- @fsolak47 for the presence reports in #2063 that put the stale neighbour recheck on the map.
- @Korenchkin for #2114, which caught the Docker instructions before anyone else did.
Every change, including the ones without a card, is in the changelog.
v2.53.0
Ten things that were quietly wrong now tell the truth, and the fleet card stops guessing.
- The status of an agent upgrade, a WireGuard transfer and the traffic of a unit all came from the wrong place: a failed agent upgrade left its download behind, the 30 d transfer was a hardcoded fallback, and the traffic chart summed every physical interface instead of the WAN (#2037, #2017, #2048, reported by @edwardally and @fsolak47).
- Graphs and the numbers under them now share the same window, and the 24 h and 7 d views carry four times the points they had (#2046, #2047, @fsolak47).
- SNMPv3 arrives, and a switch no longer reports its CPU interface as a front-panel port (#2042, #2009, @Bann-DEV).
- A unit's health score stops carrying a WireGuard penalty on routers that have no SSH access (#2041, @LemurTech).
Added
- SNMPv3 (USM) support for polled devices (#2042).
- Per-peer WireGuard counters are persisted, so the 30 d transfer is measured instead of estimated (#2019).
- The 24 h traffic chart draws 96 points instead of 24, and 7 d draws 28 instead of 7 (#2046).
- The unit information card shows the agent version and whether it is behind the embedded one (#2007).
- A guide for contributors who want to translate NetPulse into their language (#2031).
Changed
- The traffic summary below the graphs follows the range selected in the chart instead of a fixed window (#2047).
- WAN traffic comes from the WAN interfaces only, so the totals stop double counting bridges and VLANs (#2048).
- The integrations page and the README platform table list OPNsense, which the app has offered for a while (#2025, #2028).
- The offline hint of an integration no longer names RouterOS for every vendor, and the raw poll error is shown as it arrives (#2039).
Fixed
- A failed agent upgrade removes the downloaded binary and its staging file on every exit path (#2037).
- The WireGuard "Transfer 30 d" widget shows the real peer transfer instead of a hardcoded fallback, and only falls back to
wg0when the listing ran on the router itself (#2017, #2041). - A single-port access point no longer labels a port as WAN when the board declares none, and a declared WAN survives when the uplink is a bridge member (#2033, #2034).
- A MAC with no lease data stops being labelled as a static reservation while online and as expired once it leaves (#2029).
- A host with virtual interfaces is no longer badged as a switch (#2032).
- The topology map only offers the detail link on nodes that actually navigate (#2049).
- Rack profiles seeded before 2.51.0 show live ports and offer to clean the LAG and virtual ports that stayed behind (#2010).
- A scheduled config backup served by an agent logs as scheduled, like the ones the server runs (#2011).
Thanks
- @fsolak47 for #2046, #2047, #2048 and #2049: four reports that turned into more graph resolution, range-aligned numbers, WAN counters and a topology link that does nothing. The traffic screens were one person looking carefully.
- @crowedavid for #2007 and #2011, the agent version in the unit card and the inconsistent journal entry, and for the port captures behind the two agent fixes in this release.
- @Bann-DEV for #2009 and #2010, which put the switch CPU interface back where it belongs and made old rack profiles usable again.
- @edwardally for #2017, the WireGuard widget that had been showing the same number for months.
- @LemurTech for #2041, a health score penalised forever by an interface that did not exist.
Full changelog: https://github.com/gnacho/netpulse/blob/main/CHANGELOG.md
v2.52.0
Two things that failed silently now tell you the truth: Web Push, and the badge of every unit in the fleet.
- Web Push reaches iPhones again, and Settings has a Send test button so you find out your subscription is broken before an urgent alert is missed (#1974, #1973, reported by @mgrupenhoff).
- A unit's role badge follows what the unit proves instead of a config flag, so access points and NAS units stop wearing the switch badge (#1892, @Bann-DEV).
- One corrupt known_hosts line no longer takes SSH down for the whole fleet (#2014, @gnulan).
Added
- Send test button for push notifications, with the push service answer on screen (#1973).
- The WAN traffic chart draws Curve or Bars, and remembers the choice per browser. The fleet sparkline matches (#1975).
- The client type filter groups speakers as entertainment and servers as computers (#1981).
Changed
- The UI says unit for an enrolled box and client for a network client, consistently in both locales (#1980).
Fixed
- Web Push on iOS: the VAPID subject was
mailto:netpulse@localhostand Apple rejected every push with403 BadJwtToken(#1974). - Removing an agent no longer leaves the unit reporting that the agent is down (#1979).
owut checkevaluates the target picked in the UI, and a silent run says so instead of showing nothing (#1977, #1978).- A unit without a temperature sensor no longer draws a flat 0 C (#1947).
- The manual tagging form has Save and Cancel, and closing it no longer discards the draft silently (#1982).
- A unit card refreshes once an agent install finishes (#1894).
- A deleted client comes back when it returns with an IP (#1984).
- Dragging a node in the topology no longer turns into a pan or opens the tagging panel on touch screens (#1983).
- SSH survives a corrupt known_hosts entry: writes are serialized and the file is replaced atomically, and unreadable lines are dropped with a journal entry instead of being fatal (#2014).
Thanks
- @crowedavid for #1980, #1981, #1982, #1983, #1984, #1894, #1947 and #1979, eight reports that turned into changes in this release.
- @Bann-DEV for #1892, the device role discussion that ended in a badge derived from evidence.
- @mgrupenhoff for #1973 and #1974, which made push notifications trustworthy.
- @gnulan for #2014, a silent fleet-wide failure found by reading the code.
Full changelog: https://github.com/gnacho/netpulse/blob/main/CHANGELOG.md
v2.51.5
Four fixes, no new surface.
- External units created by an integration keep a bare host, so they can be edited and set as the gateway again (#1968, reported by @harpyeagle in #1056).
- A Proxmox host and its untagged guests no longer show another client's VLAN (#1757, @MassiPi).
- One unit is one SSH connection again, instead of two (#1893, @crowedavid).
- An unknown uptime shows "—" instead of a real-looking "0d 0h" (#55, @harpyeagle).
Thanks
- @crowedavid for #1893, #1894, #1899, #1944, #1945, #1947 and #1949, seven reports that turned into fixes in this release.
- @MassiPi for #1757, the VLAN work that keeps improving the panel and the topology.
- @mgrupenhoff for #1943, the iOS standalone layout.
- @harpyeagle for #55 and the OPNsense reports in #1056.
Full changelog: https://github.com/gnacho/netpulse/blob/main/CHANGELOG.md
v2.51.4
Units that will never run the agent are now first class. A router with very little flash, or one where the package simply does not fit, used to be a second-class unit: it showed up in the fleet, but its inventory was a guess, its config backup was only possible by hand and the weekly schedule ignored it entirely. This release gives it an inventory mode of its own, the config backup over SSH from the fleet list, and the same backup on schedule. In parallel, the agent that does fit small devices now ships compressed, so fewer units end up in that situation at all.
The other half of the release is about honesty in the UI. An install that took longer than a browser timeout was reported as a network failure with the reason thrown away one layer below, and a unit under plain SSH produced connection noise that no log line on the server could explain. Both are now visible for what they are.
Testing this against real hardware turned up three defects that no unit test had caught. The config backup over SSH relied on base64, which the stock OpenWrt image does not ship, so the backup failed on exactly the units it was written for. The networks view was empty on every unit, because the probe discarded the line that says what kind of section each one is. And a partial router update reset the gateway flag, so switching a unit to SSH inventory quietly dropped it from the gateway role.
Added
- Each unit declares its inventory mode: auto, agent or SSH (#1917).
- Each unit reports where its inventory came from and how old it is (#1917).
Fixed
- An agent install keeps its output, on success and on failure (#1894).
- A slow install is no longer reported as a network failure (#1894).
- The configuration backup is offered in the fleet agent list on units that only answer over SSH (#1899).
- The scheduled config backup covers units that only answer over SSH (#1924).
- The config backup over SSH works on units without
base64in the image (#1935). - The networks view lists the interfaces of a live unit, with the agent or over SSH (#1936).
- A partial update of a router no longer resets its gateway flag or its agent-only flag (#1937).
- The OpenWrt package ships a compressed agent (#1891).
- Every SSH probe that does not go through the connection pool leaves a line in the server log (#1893).
- A release reuses the agent binaries when the agent did not change (#1916).
- The networks card renders on a live server (#1918).
Discussions
- Multi tenant (#1215), Opnsense as the gateway (#1056), router integration policy (#1575).
v2.51.3
Agent self-upgrade works again. The previous release packed the agent binaries with UPX, and the startup guard that checks the embedded agent version scanned the raw bytes for a version string that compression had hidden. The server therefore started with agent upgrades disabled, so every fix that travels inside the agent was unreachable. The build now writes a small version stamp next to the binaries and the server reads it first, keeping the byte scan as a fallback for unpacked builds. Nothing else changes.
Fixed
- Agent self-upgrade works again on servers built after the binaries were packed with UPX (#1914). The #1246 guard scanned the embedded bytes for the version string, and compression hides it, so every release since the packing change started with self-upgrade disabled and logged a mismatch that did not exist. The build now writes the agent version in a stamp file next to the binaries and the server reads that stamp first, keeping the byte scan for hand-copied or unpacked sets.
v2.51.2
Your network, the way you configured it. Every client now carries the VLAN its bridge actually learned, and the router page lists your OpenWrt networks with their members, their subnets and their traffic, instead of only a chassis with jacks. The agents are a third of their former size, so an upgrade fits in a small flash again, and an agent with no init script restarts itself instead of waiting for a reboot nobody knew was needed. Plus a batch of fixes from the people who run this daily: backups that clean up after themselves and work over plain SSH, a firmware target that stays where you picked it, charts that stop inventing slopes, and a topology that finds the unmanaged switch behind your managed one.
If you run the packaged agent, this release also ships agent 3.0.37. It is required for per-client VLANs, the access point interface and the restart without an init script: a unit that keeps agent 3.0.36 will not show those fixes. On Merlin-style installs without procd the first update still needs one manual restart, and every later one replaces its own process.
Added
- The router page can now be read as the networks the unit declares, not only as a chassis with jacks (#1889). A networks card lists each OpenWrt network with its device, its subnet, its VLAN, its member ports, how many clients sit on it and the traffic of its device. Both ways of segmenting a box go through the same code: a bridge with VLAN filtering, where the network device is br_switch.999 and the members come from its bridge-vlan section, and one bridge per network fed by a tagged subinterface, where br-iot learns its VLAN from eth0.999. Traffic is the counter of the network device rather than a sum of its ports, a unit without per-interface counters shows no figure instead of an invented one, and clients are counted only when their VLAN is known. Reading the configuration also fixed a quiet parsing bug: a UCI list written as 'eth1 lan1 lan2' was being treated as one single port name.
- A configuration backup no longer needs an agent on the unit (#1899). The button only rendered for a unit with an agent, so a router managed over plain SSH had no way to keep a copy of its config even though the server can already reach it. The server now builds the archive itself with
sysupgrade --create-backup, brings it back base64 encoded over the SSH channel, always removes the temporary file on the unit, whose /tmp is RAM, and stores it where the agent backups already live, so one list and one download serve both kinds. A unit that does have an agent still prefers the event stream and only falls back to SSH when the agent is not connected.
- Reinstalling an agent now shows what the install script actually said (#1894). The handler discarded the script output with
_, so the operator got a red "exit status 1" that vanished after eight seconds and the server journal had nothing at all. The output is captured and returned with the response, on failure and on success, it is written to the server journal either way, and the UI shows it in a collapsible frame under the message. A failed reinstall no longer hides itself: the error stays until the next attempt, so it can be read, copied and attached to a report.
- The SSH pool writes what it does to the system log (#1893). A unit showed bursts of "Exit before auth" in dropbear and there was nothing on the server side to correlate them with: no host, no reason, no rhythm. Every new connection is now logged with its host and user, every discarded connection is logged before the redial that follows, and every dial failure is logged with the real cause, whether that is a host key problem, an authentication rejection, a timeout or a network error, together with the consecutive failure count and the backoff. Failures are rate limited to the first and every tenth so a dead unit cannot fill the journal, and a recovery line closes the episode.
Fixed
- An unmanaged switch behind a managed one is inferred again (#1791). The inference required a managed distribution node whose MAC matched the bridge address of the switch exactly, and it only accepted units of type managed-switch. An SNMP switch usually announces an LLDP chassis MAC that is not its bridge address, and a unit declared as external with SNMP did not pass the type check at all, so the leaf switch never appeared and its clients stayed in the fan-out of the managed switch. Every managed unit with a forwarding table is walked now, and when no node matches its MAC the leaf hangs from the fleet unit itself, keeping the physical port on each client. Access points with an agent are still excluded, because their multi-MAC port is their uplink.
- Wired clients fan away from their unit instead of always to the west (#1791). The fixed arc sent every chip towards the gateway when the unit sat on the right edge, on top of the links between units. The arc is now centred on the direction from the gateway to the unit, so a unit on the right fans its clients to the right and a unit on the left keeps the layout it had.
- Traffic charts no longer invent a ramp between two buckets (#1901). A point in these series means "the highest rate seen inside this bucket", so a straight segment to the next point claims the link spent the whole gap climbing, and a one minute burst inside an hour renders as a triangle that spans the neighbouring hours. The WAN chart, the fleet sparkline and the aggregated port ranges now draw steps, which say exactly what was measured and no more. The 24h port chart keeps its straight segments because it reads raw instantaneous samples, and the latency chart is untouched since it is not a max-per-bucket series. One detail differs from the report: every WAN range is bucketed, including 1h, whose points are three minute maxima, so the step applies there too instead of only to the hourly view.
- Discovery no longer offers a unit that is already in the fleet (#1895). A candidate was marked as already added only when its IP was byte-identical to the host string in the configuration, so a unit added by DNS name, by a name with different case or a trailing dot, or by an IP with a port showed up with an Add button while the gateway, added by plain IP, was the only one recognised. The comparison now normalises case, port and the trailing dot, resolves configured names to their addresses, and also matches the reverse DNS name of the candidate, with name resolution bounded so a stale name cannot slow the scan down.
- A config backup no longer stays on the unit after it is uploaded (#1897). Every manual or scheduled backup wrote its tarball under
/tmpand left it there, and on a router/tmpis RAM, so each snapshot took memory away until the next reboot. The agent now deletes the local file once the operation ends, whether the upload succeeded or not, since the copy that matters is the one the server keeps with retention. The firmware flow keeps its local backup on purpose: that one is reported in the result so it can be recovered by hand when sysupgrade fails.
- The target version chosen on a firmware card stays where it was left (#1898). The selection lived only in the local state of the page, so a refresh, leaving and coming back, or the ten second auto-refresh that runs while an upgrade is in flight put the dropdown back to whatever the server had stored, and nothing was saved unless the separate Save button was pressed. Choosing a version now saves it, a reload no longer overwrites an edit that is still local, and scheduling a recurrence saves the target shown in the dialog before saving the schedule. That last case mattered beyond the dropdown: a scheduled run flashes the target stored on the server, and a unit with none had its recurrence disabled when it fired, with "no firmware target saved" as the only explanation.
- A client is tagged with the VLAN its bridge actually sees (#1757). The forwarding table was read from one bridge only and every port carrying a dot was thrown away, so a router that segments with 802.1Q subinterfaces, where the real ports are
eth0.777andeth0.999, lost its wired clients altogether: no port, no VLAN and no line in the topology. Every bridge the unit has is read now, a subinterface counts as a port, and each MAC carries the VLAN the bridge learned it on, which is the only evidence of where a client lives. A forwarding table without that field leaves the client without a VLAN instead of guessing one from the default of the port. - A wireless client keeps the interface and the SSID that served it (#1757). The hostapd path built each station without either, so on any OpenWrt that answers over ubus every WiFi client arrived with an empty interface, and the topology, which resolves the VLAN of a station through its AP interface, had nothing to colour. The interface is taken from the object that was queried and the SSID from
iw dev, and the VLAN of an AP interface now carries the name and the subnet of the network it serves. - An upgrade that swaps the binary but cannot restart says so instead of leaving the unit lying (#1900). On a host without a procd script there is nothing to ask for a restart, so the new binary stayed on disk, the old one kept running from memory, and the UI reported the unit as up to date until the next reboot. That covers the Merlin-style installs under
/jffs, Synology and any manual or Entware deployment, which are exactly the units that ended up as external in 2.49.x. The agent now replaces its own process with the binary it just installed, keeping the PID, the arguments and the environment, which starts the new build in place and leaves the OpenWrt path untouched. If even that fails, it reports the failure with the reason while it is still alive. The availability flag now uses the same criterion as upgrade all, a live agent stream rather than the row type, so the badge and the command stop disagreeing. And a restart that never completes stops sitting at restarting: after ninety seconds the row shows a terminal failure that names the cause and lives longer than the ordinary progress state, so an operator sees it instead of an empty slot. - A guest without a tag is no longer painted with a VLAN it does not have (#1757). When Proxmox declares no
tag=, the guest sits in the native VLAN of the bridge of the host, not in the VLAN of the port where the host is plugged. Inheriting the VLAN of the host chip marked those guests with an unrelated network, so a container on an untagged bridge showed up in the VLAN of a segmented one.
Changed
- The link out of an alert reads View unit instead of View router (#1896), which is the wording the rest of the interface already settled on. The device list link was renamed with it, and the Spanish strings say Ver unidad.
- An agent ships compressed, so a router with little room can install it (#1890). The update downloads the new build to
/tmpand moves it over the installed one, which means a unit needs twice the size of the agent free before it even starts: 15 MB for a 7.5 MB agent, on a flash of 16 MB. The report came from a unit with 10 MB where the upgrade could not begin. Every architecture is now packed with UPX in--best --lzma, which takes the five binaries from 6.9-8.3 MB down to 2.0-2.4 MB, and the cost is about 170 ms at start-up on x86. The build refuses to produce agents without UPX instead of skipping the step in silence, a test fails when an embedded agent is not packed, and the release checks the mark and the size before publishing. The embedded sha256 is computed over the packed bytes, so integrity checks are unchanged.
Commits
- 70361cfb8c9ef688de8c9efb9f5b9abc37e66810 build(agent): pack the agent binaries with upx (#1890) (#1902)
- 65393f182eb1f4eb591d7803bfc97fb4084d16c9 chore(agent): bump the agent to 3.0.37
- d4c419bea8c69e58ce117cd711fa17d9ba8facc1 chore(release): 2.51.2
- 793e268a5495af31ab878cb33ae327fb2333897a feat(agents): return the reinstall output instead of only a red banner (#1894) (#1910)
- 25c174b40bcaa4864e507ca4ccedb0209bd9569e feat(backups): take a config backup over ssh when there is no agent (#1899) (#1911)
- 5c9c44e383e7a09bcbda7adf4bfc71a4b88d81bc feat(routers): read the unit as openwrt networks, not only as a chassis (#1889) (#1912)
- 0143a2eb2627abaefedbc5ad4fc9c19c30966845 feat(ssh): log every dial, discard and failure with its reason (#1893) (#1907)
- 84100e713fa0a6537569fd96f721aaa5436cbdc0 feat(wireguard): hide the peers that are not connected (#1886) (#1887)
- dff6ce1acb06aafbf1b88aaaa21441b74049dcd5 fix(agent): delete the config backup from the unit once uploaded (#1897) (#1905)
- 2b5467e796e911c97881d569c1cdfb580c2190d4 fix(agents): close the upgrade cycle on hosts without a procd script (#1900) (#1903)
- 3288e0587e9f038b71743d08dc8e49266db8a6b6 fix(charts): draw aggregated buckets as steps, not slopes (#1901) (#1908)
- c92ad3ee193feb96a48396a8b50f052cdb5c7029 fix(discover): recognise fleet units that are not added by plain ip (#1895) (#1906)
- 525ec0975866a2a70232eaf5646de0a1db3a57e4 fix(firmware): keep the target version the operator picked (#1898) (#1904)
- 917d05e1b7cbcf72ac3d2d7e42514c1ed285acb2 fix(topology): infer the leaf switch behind any managed unit and fan clients outward (#1791) (#1913)
- 700b254de99588a0683457287df52569b17f4ef5 fix(ui): say view unit instead of view router (#1896) (#1909)
- ad058b89b3cbc44e97adf0b7e064ce7fd65164da fix(vlans): tag each client with the vlan its bridge learned (#1757) (#1888)
v2.51.1
Changelog
- eb84c457ee2e98a45c898b714b8f91683c1517c2 chore(release): 2.51.1
- bc3bacd845417af126ea0bdaf2725f19fa035318 docs(changelog): cover the changes since 2.51.0
- e400783f45504ab7f4e47fc32b4d0fec713556eb docs(changelog): credit the 2.51.0 contributors
- e5e725024341eaf5ad2e2445b08baa3e0cf2a8a7 feat(channels): draw the live lens from scans when the unit has no survey (#1878)
- 7c4e26648a43ec529dd75fd5610303a56f308ae9 feat(devices): leave a low-priority note when an unverified MAC persists (#1877)
- de6df7faad19682cddeea96e1234e244a9720ef6 fix(agents): keep the agent-router association on wireless-only pushes (#1875)
- 55825c5d82446bae39348784b47a2b183c2a52ee fix(agents): let upgrade-all count a live agent stream as eligible (#1883)
- 3050318e8e174d4815a2f7f5466fe26f3646795a fix(agents): send the install script over stdin so dropbear accepts it (#1884)
- 0e2f652c98d34df3877a834d943a15eeec5a6132 fix(devices): only alert on unnamed MACs with evidence of a host (#1818) (#1869)
- 3d776402abac6884e5df1abaaa7e90ea4d983918 fix(pve): recreate the client when the TLS settings change (#1880)
- dc2223bc43fb59c5eba0e931e09fd0724e2c8b1c fix(routers): fill the 24 h line from port series for pusher-only units (#1876)
- 0d3c6185d6c52fea68fb50b4d5bf5d0b2b3d723e fix(update): refresh a stale check before applying (#1882)
- 66e7488b9d5fa764331c9cfb737b6a9eb335e274 fix(vlans): read the VLANs a router and a hypervisor declare (#1757) (#1885)
v2.51.0
NetPulse v2.51.0
Two months of work since v2.50.0: 86 pull requests, 88 commits and 58 documented changes, most of them reported by people running NetPulse on their own hardware. This release is as much theirs as ours.
If you run the packaged agent, this release also ships agent 3.0.35 and the server embeds it, so a reinstall, a self-heal or a package download from the UI brings every unit up to date.
The short version
- Backups are now useful off the machine they came from. A nightly backup is a single
.tar.gzwith the database, the.sshdirectory andtls/, taken from a consistent snapshot. Restoring on a fresh host brings back data, the SSH key the server uses to reach the routers, and the certificates. - A reinstall can no longer leave an agent down, silent or downgraded. The installer says which step failed, keeps the service running until it has something to install, finishes the setup even when a step fails, and refuses to replace an agent that is newer than the binary the server carries.
- Agents come back in seconds after a server restart. The SSE reconnect delay used to grow to five minutes on every restart because an unclean cut counted as a failure.
- Short-lived addresses are no longer clients. A MAC that comes and goes within a minute never reaches the device table, the counters or the "new client" alert.
- The channel plan stops suggesting moves that change nothing, and when a radio refuses to scan it now says why instead of showing an empty table.
- Less disk, for longer. Five-minute buckets are kept for 90 days instead of a year, and the copy the updater takes before an update is compressed, which takes a 419 MB database down to about 130 MB per copy.
Upgrading from v2.50.0
These are the behaviour changes worth knowing before you upgrade. Everything else is additive.
| Change | What to expect |
|---|---|
| Backup format | New backups are netpulse-<timestamp>.tar.gz instead of .db.gz. Restore with tar -xzf netpulse-<timestamp>.tar.gz inside the data directory. Existing .db.gz and .db files are still purged by the retention policy, they just are not produced anymore. |
| Backup contents | The archive holds credentials: netpulse.db, .ssh/ and tls/. Treat it like a secret. |
| Five-minute retention | client_bw_5m and port_series_5m now keep 90 days instead of 365. Raw samples stay at 7 days, daily rollups are kept forever. The first purge after the upgrade will delete 5m rows older than 90 days. |
| Update copies | deploy/update.sh compresses the pre-update database copy. Rollback restores from the .gz and drops stale -wal/-shm files first. |
| Agent downgrade guard | The binary, package and reinstall endpoints return 409 agent_downgrade_refused when the unit reports a newer agent than the one embedded in the server. Add ?force=1 to override. A unit with no reported state is never blocked, so a dead agent can still heal itself. |
| New client dwell time | A MAC unknown to the server must be seen for at least 60 seconds before it counts as a client or fires an alert. The waiting window lives in memory and starts over when the server restarts. |
| Per-unit push interval | The agent cadence is a per-router setting (agent_interval, 0 = the agent default of 30 seconds). It is written to the UCI config on routers that carry the package. |
Added
- The ASU exclusion list can be written on the router from the firmware card. When
owut checkreports packages missing in the target version, the card offers to append them toattendedsysupgrade.owut.remove, so the next build keeps them. - The exclusion list is read from the router itself.
owut checkalso reads the list the owner already maintains, instead of only the one NetPulse wrote. - The server can serve the OpenWrt agent as an installable package. It builds the
.ipkon the fly from the embedded binary and serves it from/api/agents/{slug}/package, so a reinstall goes throughopkg/apkand lands in a conffile that survives a sysupgrade. - The agent push interval is a per-unit setting. A busy access point wants a short cadence, a modem on a metered link does not.
- A radio with no neighbour scan now says why. The agent keeps the last line of
iw's stderr and sends one outcome per radio with the interface it ended up using and the number of BSSes it saw. The channel plan shows it under the radio. - The firmware card can reset the upgrade state of a router. An attempt stuck in
runningorrebootingfor more than fifteen minutes no longer needs an API call to clear. - Clients in the topology show their VLAN, wired and wireless. The router list carries the bridge VLAN filtering table and each chip is coloured by the VLAN of the port it sits on.
- Configuration backups can run on a schedule. A card on the Firmware upgrades page picks a weekday and a time, and the server keeps the last and next run so a restart does not lose the schedule.
- Pending upgrades show their configuration backup. The upgrade card exposes the archive that belongs to the unit, so restoring after a flash no longer means hunting for a file by date.
- New firmware alerts. NetPulse now raises an alert when a new firmware version is available for a device, with the target version in the detail.
Changed
- An automatic backup is a single
.tar.gzholding the database, the.sshdirectory andtls/. It used to carry onlynetpulse.db, so restoring on a fresh host brought back data and configuration but the server could not reach a single router until the key was recreated by hand. - Five-minute buckets are kept for 90 days instead of a year, and the pre-update copy is compressed. On a live install the database had reached 419 MB, 182 MB of it in 5m buckets built in 64 days, and an uncompressed copy per update did not fit the disk.
- Every page shares the same width and the same header gap. Alerts and Settings capped their content at their own width while the rest used the shell, so the pages looked like different products.
- Every page header uses the layout of the topology page. Breadcrumb in small caption with a chevron, one-line explanation below, actions on the right. This touched Home, Alerts, Devices, Settings, Channels, Reports, Firmware upgrades and Help.
- The page title is no longer printed twice. The top bar already shows it, so the copy under the breadcrumb is gone.
- The per-unit actions live together. Regenerate token moved from Settings > Fleet to the Actions column of the agents card, and Forget host key moved from the unit detail to the same place.
- The in-app help covers integrations and certificates. Two flows are new: the add-ons group in Settings and the vendor catalog, and the certificate and plain HTTP options.
- The rack picker names its groups the way the rest of the UI does, and the rack editor grew a set of long-requested fixes:
- a unit can be mounted at full width, half width or a third of the rack, so two or three devices fit on the same shelf;
- a rack is created with any height from 1 to 45 U instead of six presets;
- ports of a racked unit show their name instead of the SNMP ifIndex;
- patch panel ports are numbered 1 to 48 instead of the internal
p01key; - the picker favours the unit name over its model, and lists every wired client that can be racked;
- the front panel only shows physical ports, not the VLAN interface or the LAG.
Fixed
Backups and retention
- An automatic backup no longer fails with
archive/tar: write too longon a live server. The archive sized the database entry from astatand then copied until EOF, so a database that grew between the two calls wrote more bytes than the header declared. It now takes aVACUUM INTOsnapshot first and copies exactly the declared size. - The nightly backup no longer leaves the SSH key and TLS material out.
- Update copies are compressed and the rollback restores from the compressed file.
- Five-minute buckets stop eating the disk (90 days instead of a year).
Agent, installer and self-heal
- A reinstall that fails no longer leaves the agent down. The installer stopped the service before it had anything to install, so a download that could not reach the server or a binary whose checksum did not match left the unit stopped with the old binary in place.
- A reinstall that cannot finish its setup says so and still brings the agent back. After the binary swap, any failing step of the remaining setup aborted the script with the service still stopped. Environment file and init script are now written to a temporary file and moved, and the restart is always attempted, with a distinct exit code when the agent does not come up.
- A reinstall that cannot download the agent binary or install the package now says so on stderr, instead of dying silently on the first failing command.
- An SSH reinstall that fails reports the last line the router printed on stderr, not just
Process exited with status 1, so the reason (a failing curl, a full filesystem, a missing command) reaches the operator. - The init written by the installer no longer registers a service whose binary is missing. After a sysupgrade, when the self-heal could not fetch the agent, procd was told to run a path that did not exist and the unit stayed dead until someone looked.
- A reinstall no longer downgrades an agent that is ahead of the server.
- An agent whose SSE stream is cut by a server restart reconnects in seconds instead of minutes.
- Rotating the agent token now reaches the configuration the agent actually reads. The packaged init prefers the UCI config and only falls back to the environment file when the UCI one is empty, so the rotation used to leave the old token in the file that mattered.
- The served installer is always LF. A checkout with
core.autocrlf=trueput CRLF in the embedded script and busybox refused to parse it. - The packaged OpenWrt agent now receives its UCI configuration. The procd service passes server, token, slug, interval, scan interval, ping targets and SPKI fingerprint in a single call.
Clients and devices
- Addresses that come and go within a minute are no longer clients. Randomised MACs, phones that probe once and mesh units that bounce an association all reached the device table, the counters and the "new client" alert.
- Locally administered MACs are recognised by the server itself, not only by the agent discovery scan, so a MAC that lives in the switch FDB is flagged too.
- A locally administered MAC with no address and no lease is no longer remembered.
- Units report the MACs of all their interfaces. Only the bridge MAC was absorbed, so the CPU port of a DSA switch, a VLAN interface or a radio could be learned by a neighbour and appear as an unknown client.
- The default route of the gateway no longer appears as a client.
Channels and topology
- The neighbour scan survives a radio whose AP interface refuses to scan. On Asuswrt and other Broadcom firmwares the AP interface answers
-95while the managed one works. The agent now retries on another interface of the same radio, at most twice, and reports the outcome. - The channel plan stops moving a wide radio to the other half of the channel it already uses. The current block was matched by equality, so a 40 or 80 MHz radio on a non-primary channel matched nothing, scored zero and got a suggestion that was the same block.
- Opening the Channels page no longer starts a WiFi scan with the periodic scan off.
- Topology links end on the client node, clients behind an inferred leaf switch hang from it, and right-edge chips fan to the right instead of piling over the incoming links.
Monitoring and SNMP
- A managed Netgear switch reports its chip again, reading model, chip and firmware from the FastPath enterprise subtree.
- TP-Link switches report model, firmware, CPU and memory through their enterprise MIB, since the firmware answers neither ENTITY-MIB nor HOST-RESOURCES-MIB.
- SNMP switches read the generic ENTITY-MIB, so a monitored switch shows its firmware and a sensible model name.
- MikroTik temperature comes from the health table, with CPU and RAM from HOST-RESOURCES.
Firmware and owut
- An owut upgrade that finishes updates the stored firmware target, so a unit upgraded through the flow stops showing the old version.
- The owut flow is two steps.
installruns without an image and the download gate cannot drop third-party packages. - The firmware target dropdown defaults to the latest version and lists what ASU would build, instead of a stale
owut versionslist. - An owut install no longer marks a successful flash as failed when the router reboots.
UI and settings
- A gateway without AdGuard no longer leaves half a row empty. The row width was decided by the visibility checkboxes while the card decided on its own whether to render.
- The pre-flight checks speak the language of the UI. Their detail text was written by the server in Spanish, so an English interface showed
3393 MB libresnext to a translated label. - A resolved alert shows what it resolved, instead of a bare internal id.
- The settings group goes back to its Integrations label, and the section that hosts the vendor integrations manager is the one called Complementos (Add-ons in English).
Thanks
This release exists because people took the time to file a good issue, attach a log, try a build and tell us it did not work. Every name below changed something in this codebase.
@Bann-DEV reported ten issues and sent five pull requests, including the Netgear FastPath chip and identity reading from its enterprise MIB. His reports covered MikroTik health data, pusher-fed units that were never polled, the roaming tile that was always zero, per-port rates without a traffic line, 6 GHz recommendations, pusher visibility on the Integrations page, interface MACs showing up as unknown clients, ENTITY-MIB, the UCI config that never reached the process, and the CRLF installer. Several of the fixes in this release exist because he kept testing after the first round. Issues #1725, #1726, #1740, #1741, #1742, #1743, #1765, #1767, #1772, #1812.
@crowedavid filed eight issues that shaped this release: the resolved alerts that lost their detail, the traffic card that smoothed too much, three rounds of topology work, the Channels page that scanned on its own, the backup configuration feature, the SNMP vendor quirks design, the fleet agent and SSH key management review, and the rack diagram enhancements. He also pushed back on a closed issue and made us look again. Issues #1466, #1786, #1791, #1794, #1795, #1802, #1803, #1804.
@MassiPi asked for custom SSL certificates and reported the interface, VLAN and MQTT problems that turned into the VLAN chips in the topology. Issues #1756, #1757, #1764.
@jeffarndt caught the pre-flight checks answering in Spanish while the UI was in English. Issue #1811.
@edwardally asked how to change the router temperature alert threshold, which is how the alert configuration path got reviewed. Issue #1736.
And to everyone running the development builds, restarting their router to test a fix, and pasting the log when it still failed: thank you. That feedback loop is the reason the installer and the agent are much harder to break than they were two months ago.
Full changelog: CHANGELOG.md
Compare: v2.50.0...v2.51.0
Install: curl -fsSL https://raw.githubusercontent.com/gnacho/netpulse/main/install.sh | sh
v2.50.0
Router integrations are here. This release completes the Integrations section that deploys, runs, updates and edits vendor pushers for you: MikroTik RouterOS, Asuswrt and Asuswrt-Merlin (installed on the router itself over SSH, the Merlin way under /jffs or with a password sign-in on stock dropbear), UniFi controllers, and a vendor catalog that grows without touching the core. The integration code lives in a separate repository, netpulse-integrations, and each vendor ships as its own standalone binary speaking the same signed ingest contract as the native agent. You only ever download and run the code for hardware you actually own: no UniFi controller, no UniFi code on your box. That is the deal that makes NetPulse compatible with more platforms every release while the server binary carries no vendor code inside and the shared code stays a small contract anyone can review.
Added
- UniFi controller vendor (#1695): a read-only
unifi-pusherfor self-hosted UniFi Network controllers, both UniFi OS consoles (UDM, UDM Pro, UNVR and similar) and classic standalone ones; ui.com cloud accounts are not supported, the pusher only speaks to a controller reachable on the LAN. One site per integration: gateways, switches and access points with their ports and uplinks (the topology hangs together without LLDP), wired and wireless clients with band and signal, MAC-to-IP mapping, and gateway uptime and identity. Surfaces with no verified read-only match (DHCP leases, per-interface counters, VLANs, roaming, CPU, temperature) are not emitted. It deploys from the Integrations section like RouterOS, on the host that runs the server. - The Integrations vendor catalog ships as a manifest (#1706):
vendors_seed.jsonnames each vendor with its release tag prefix, binary and architectures, so adding a vendor or an architecture is a manifest entry instead of a compiled table in the core. - Integrations in ssh mode accept password authentication (#1708): when the optional companion file
<data-dir>/integrations/<slug>.sshpass(0600, same discipline as the environment file) exists, the server signs in to the router with the integration username and password; without it the server key stays the default and preferred path. The create and edit forms reuse the same password field for it (empty on edit keeps the current one), the password never reaches the database or any API response, deleting the integration removes the file, and an authentication rejection reports its ownssh_auth_failedcode instead of the genericssh_failed. Stock Asuswrt (dropbear with password auth only) now works with no manual key setup. - Inactive agents in the fleet table now offer Rearm, Reinstall and the copyable install command next to Remove, so a rejected token (HTTP 401 after a reboot or a server move) can be fixed without removing the agent or losing its history. NetGrip rows offer Rearm only and external pushers keep Remove only.
- The card and the detail of a unit pushed by an integration (RouterOS and other vendors) show why it is offline: authentication rejected, missing API endpoint (old firmware, REST needs RouterOS 7+), connection refused or pusher stopped, with the raw pusher error on the tooltip.
- The on-box server init script now refuses to start when its data dir is not a real mount (default overlay paths excluded), so an unmounted USB stick can no longer leave a fresh SQLite database on flash. It also moves to
START=99to let USB mounts settle first. - The agents table marks a rejected-token agent with its own badge the moment the pusher authenticates with a stale token (a reboot or a server move does it), and the badge clears itself on the first valid push. The Integrations rows carry the same offline classification as the unit detail, so a dead integration explains itself in place (#1719).
Changed
- The manual documents the agent recovery flow (Rearm restarts the agent service, Reinstall rotates a rejected token) and the agent tooltips and banners point to the agents section on Devices, which hosts the reinstall action, instead of the router detail page.
Fixed
- Integrations in ssh mode on procd vendors (#1709): the generated init now loads the router environment file into the pusher process with a single
procd_set_param envcall (extra calls overwrite each other, so only the last variable survived), refuses to start when the environment file is missing or unreadable, and keeps the file watcher so an env change still restarts the service with the new values. Before, a procd vendor started its pusher with noNETPULSE_*or vendor variables at all. - Heavy write volume on-box (#1700): the on-box server now runs SQLite in WAL mode with
synchronous=NORMALandtemp_store=MEMORY, the same engine as every other deployment. The previous on-box choice (rollback journal DELETE + synchronous FULL, Fase 9 R6) rewrote pages and journal on every transaction and tripled the measured write volume (45 GB/day on a USB stick with a 107 MB database, versus 13.7 GB/day in WAL with the same fleet). WAL withsynchronous=NORMALcannot corrupt on a power cut; only the commits not yet synced are rolled back. The hourly checkpoint and nightly jobs now run for on-box installs too, so the-walfile stays bounded. - SQLite aggregation spills (the
etilqs_*temp files) no longer hit disk by default. The report measured two ~3.16 MB temp files per 60 s window, about 9.1 GB/day and two thirds of the server's process writes with a ~100 MB database; withtemp_store=MEMORYthe same aggregation happens in RAM (a few MB per query). - Asuswrt over SSH installs the Merlin way (#1705): binary and environment under
/jffs/netpulse, a small pidfile control script for start, stop and status, and the boot hook as a managed block inside/jffs/scripts/services-start(removed cleanly on uninstall) instead of a procd service, which Asuswrt does not have. Nothing is written outside/jffsexcept the hook line. - Enabled local integrations resume on boot (#1704): after a server restart every enabled pusher starts again, and rows left stale by a crash or a kill reconcile to what is actually running.
- Pre-pin pusher environments heal on update and edit (#1699): an integration created before the SPKI pin change gets its environment rewritten by the next update or edit instead of staying pinned-less until it is deleted and recreated.
- Pusher environments ship the server SPKI pin (#1685): external pushers verify the server TLS certificate against the pinned public key, the same trust model as the native agent.
- The greeting line on the Overview ("N important alerts") now counts the same alerts as the bell and the Alerts page (unread, level-filtered), instead of the health score breakdown. A health penalty without a visible alert, such as a down VPN peer with the vpn category set to "Nada", used to leave a phantom counter that could not be cleared from anywhere (#1707).
- OpenWrt 21.02 on-box package (GL.iNet GL-MT6000 report): the
.ipknow installs/etc/init.d/netpulse, the agent init and the uci-defaults scripts as executable (0755), and the packaging scripts verify the modes inside the built.ipk. - OpenWrt 21.02 on-box package: the procd init now uses ONE
procd_set_param envcall with all variables (in 21.02 each extra call overwrote the previous one, soNETPULSE_ONBOXandDATA_DIRnever reached the process), andstdout/stderrare set as boolean params so server logs show up inlogread. - Gateway autodetection no longer picks the ISP-side (public or CGNAT) address on an on-box server: when the default route's via is not a private LAN address, the gateway is the router itself and its LAN IP is taken from the private subnet route. Dedicated LAN servers keep the previous behavior.
- The server generates its SSH keypair in-process (crypto/ed25519 + x/crypto/ssh) instead of calling
ssh-keygen: on OpenWrt 21.02 the available keygen is Dropbear's and writes a format crypto/ssh cannot parse, which silently degraded the server to the demo dataset. An existing unparseable key is backed up and regenerated, and the public key fingerprint is computed in-process. - The default on-box port is now 3100 instead of 3000, which collides with the AdGuard Home UI shipped by GL.iNet firmware. Existing installs keep their configured port.
- The initial bootstrap password is removed from UCI after a password change, so the generated plaintext no longer stays in
/etc/config/netpulseforever. - README:
DEMO_MODE=1requiresAUTH_PASSeven in demo mode, and the settings section holding the SSH public key is called Fleet (formerly documented as Network). NETPULSE_DB_ROLLBACK_JOURNAL=1restores the previous on-box SQLite journal engine (DELETE + synchronous FULL) after the #1700 default change;NETPULSE_DB_TEMP_STORE(memory|file|default) controls where SQLite aggregation spills go, and is also settable on-box via the newdb_rollback_journal/db_temp_storeUCI options.
Security
- Dependencies:
minimatchforced to 11.0.3 through overrides (#1721, Dependabot alert 37). The globparts amplification advisory has no 10.x patch and the package only reaches the build through the lint toolchain, so the override is the only path. - Dependencies: pick up the gonzalop/mq audit fixes (#1693).
Full changelog: https://github.com/gnacho/netpulse/blob/main/CHANGELOG.md
v2.49.4
Changelog
- 261d001833361443c1c8e34612af7dcfd2ca8dcd chore(release): 2.49.4
- 735d5a52c4f29a4bf77d031b14b2d9b870774767 docs(contributing): require a recap of open issues before cutting a release (#1688)
- cfd78b6396ca7bd00154c9715d4c70301ab7f85a docs(manual): document the pinned client filters and the remaining filter controls (#1690)
- 6fcbe371f1523030671e6417f11cb84cd3a49000 feat(devices): pin the client filter set as the view default (#1689)
v2.49.3
Changelog
- 2bd7ce882239773f0a9a6dc8b1398a05788c8dba chore(release): 2.49.3
- 8110db33bf6a14adeaefdb73dea77b0b6dae4a91 chore(release): 2.50.0
- de3859d6f49a09c2e640eacba288cfdfdee3e615 docs(readme): mention the reverse DNS fallback and deploying RouterOS from the app
- b8e2e9128d77318bc72d751fc607fd69283c7a4c feat(devices): fall back to reverse DNS for unnamed static hosts
- bf8cea80d6303bc6c710fc0a0f93af5b40499b88 fix(devices): recompute the type and unmapped address for PTR names
- e5eb36e96cd1fb7f922e917f45c8519c054f99d7 fix(topology): resolve the fleet root port from the elected parent, not only the gateway MAC (#1683)
v2.49.2
Changelog
- 25753806d2f026ffb0ce8e84898c678191d31b41 chore(release): 2.49.2
- ee9420a8f1efa893b921a9e43d4fe41996637632 feat(integrations): let an existing integration be edited without deleting it (#1680)
v2.49.1
Changelog
- 276d4188ddb66e6f37096c25ff47c930b68cbef7 chore(release): 2.49.1
- 5f4151c69247caa4fdd648813e79102888f15ccb fix(integrations): write the pusher interval as a Go duration (#1678)
v2.49.0
Changelog
- 2fbfdbb6971f296314a45973eba93c9b87c106c0 chore(app): move build-only PWA dependencies to devDependencies (#1675) (#1676)
- a7ef55db894b769fff35df65b19ca83a929ce63d chore(release): 2.49.0
- a9fcdb2d77398527270cbaabcf01c7235e4e0bae feat(integrations): download, run and watch external router pushers from the app (#1677)
v2.48.6
Changelog
- b97f772a4ab86b7f9ea1130293e920a1cf887051 chore(release): 2.48.6
- 2373ef4bbcc103268425f9acffd2c93a596bf742 fix(alerts): align frontend default config with the backend (#1672) (#1673)
- 4200b8bc7540596f3185e67a65c0f854017b1c60 test(httpapi): retry the HTTPS test listener instead of swallowing the bind error (#1671)
v2.48.5
Changelog
- 3ef2da07655d64f8bce0737cbfc6b5f212e0f1e6 chore(release): 2.48.5
- 991e07670772ff9737a2d545a7f82f8952b15655 fix(alerts): filter the built feed so demo canon events honor the levels (#1662) (#1670)
- abeecb4769ff9ced283ba9779c89c6ffe3f4ba69 fix(landing): stop the code box and feature cards from overflowing on small screens (#1668)
v2.48.4
Changelog
- a8ea80fe38ec019ef2f15ba7bac401e5b77592f7 chore(release): 2.48.4
- 0fe33d2ddd607d8b296a01b84ba0a3b6246fb0b7 fix(alerts): hide muted categories from the recent alerts card (#1662) (#1667)
v2.48.3
Changelog
- 816028f8a1a3ca38855545f7519bb73e2fa1bd71 chore(landing): bust the i18n.js cache after adding the integrations page (#1666)
- 139590c2080e969859c33d0849e0e6c0a7ed7506 chore(release): 2.48.3
- 3883b232e68bb1878d52475881271b2894c0e60a feat(deploy): publish the demo from the release artifact with a timer on the demo host (#1657)
- dded1cc4fc4a3a096d9227e461c6073cf5ec901f feat(landing): add an integrations page for the MikroTik and Asuswrt agents (#1661)
- 8570a30bfd020b7ab532c48736bcc5dbb00d7650 fix(alerts): hide categories set to none from the feed, bell and log (#1660)
- c978cfa58466770ad4d434c5dabc9b1d32b47f16 fix(alerts): keep the demo feed in line with the category levels (#1662) (#1665)
v2.48.2
Changelog
- 88065ec7af34070c654f87e487454ad5b6743175 chore(release): 2.48.2
- 4eaa5e83d157a5f5ee74a56a93e7e8a3e77446ba feat(agent): Broadcom wl tier for wireless clients, radio bands and channel width on dhd (#1531)
- 0d0d1cde07be45b1def94ebf9d1ad254bca04238 fix(agent): drop Broadcom internal ghost interfaces from the port panel (#1539)
- 41eaefa2fae31639fb72cbde3f5d0341e6c40fe6 fix(install): preflight the server URL from the router and copy the binary atomically (#1655)
v2.48.1
NetPulse 2.48.1
Fixed
- Restores LAN access for existing installs that upgrade to 2.48. The HTTP listener no longer binds only to loopback when TLS is not enabled. Existing installs keep listening on all interfaces, fresh installs stay on loopback, and the choice is remembered in
data/.bind-mode. - If you upgraded to 2.48.0 and cannot reach the UI, upgrade to 2.48.1. Immediate workaround on 2.48.0: set
NETPULSE_BIND=0.0.0.0in the service environment and restart.
Changed
- Refreshed the app and Go module dependencies.
- Pinned GitHub Actions to commit SHAs and verified downloaded CI artifacts.
- Includes agent 3.0.28 binaries.
Upgrade
Download the netpulse_2.48.1_linux_amd64.tar.gz asset for the server, or use your usual install method. The agent binaries are published as netpulse-agent_2.48.1_linux_*.tar.gz.
Closes #1647.
v2.48.0
Changelog
- 2212f7a4506274b01443a9469b19c5eea0570306 build: add make check mirroring pr-checks (#1585)
- d12a854a9f96042fe837271076bcdf67fe1d7977 chore(agent): bump version to 3.0.27 (#1597)
- 544b5a12283a2cca7ddd888a9dbf50a4ab1a6296 chore(deps): bump fast-uri from 3.1.8 to 3.1.9 in /app (#1552)
- d498778db077d4c9b9859eb53aa3b87242b38b5a chore(deps): bump serialize-javascript from 7.0.7 to 7.1.4 in /app (#1553)
- 7d55d4df80d4ac74d1787ee804438a1ec5e2891b chore(deps): bump zod from 4.6.5 to 4.6.7 in /app (#1551)
- c341da35065be695beac624bcf325f87c1e82d34 chore(release): 2.48.0
- 3a63868b60bfc43e1049cd0a9706b3eab5781d64 chore: remove internal working artifacts (#1583)
- ce971c7fb463efb1a837311b5998a565acade0c7 ci: run build and tests on every pull request
- bee7dbd6690048f26f5608e86dfdbe4866976f57 docs(contributing): hardware support bar and evidence
- 642faef2cb6aaee1394618786f18c3e2d7120a13 docs(readme): asuswrt/merlin manual install and platform table (EN+ES)
- 946ca06da62dd860e4009885bc659207e0826e65 docs(readme): generic thanks in the README, per-release detail in the changelog (#1514)
- 077232050b8287759a07a7c31137c5a96972a8d1 feat(alerts): collapse weak-signal closures and add hide-recovery setting (#1522)
- 362b6cf700bf0fcfe026065dec16eaf7e72003a1 feat(beacon): record octet counters from the RTLPlayground beacon v7 (#1536)
- 0b0c4ec11afda103f4dd1e9687febb4dc9d72416 feat(beacon): record per-port error counters (#1620) (#1622)
- aa89bdb96fbb2f0126e18e5d105d165b7800312e feat(devices): attribute SNMP switch-port traffic to FDB clients (#1554)
- b92245edbc3833ecfc5a4ed4a393cc06fdf27986 feat(devices): attribute switch-port traffic to SNMP clients for local flows
- 153d034fce57441f07da59156626418e340b7252 feat(snmp): CPU and temperature for Netgear FastPath switches (private MIB 4526) (#1528)
- 210b2924c6394c1aa4874f9c417a5e913947ea8e feat(snmp): Netgear FastPath memory + make switch vitals actually display
- ba8e1ba5feb4f297dc703d7a57f771db391513c4 feat: make neighbour scans on demand by default (#1564)
- 2c0a01d132caa7301e94e9a1bb543f5ed2b5f3a0 fix(adapters): forget host key clears entries for custom SSH ports (#1525)
- c825ed58b9d2d323c6d43c8f1c5bc0210f5509bf fix(adapters): guard lastPolled reads with the mutex (#1592)
- c740df57ba3f7f71359f15822a6fb7671a7a89a9 fix(agent): bridge FDB on Asuswrt/Merlin, wireless-side hint for rotated MACs (#1513)
- 82cc4e99969fe8be9f34049e6e737954b6c79745 fix(agent): pass UCI snapshots via stdin (#1618)
- 1a2cb21ae196af2397bcfd79fbf56c5009b35645 fix(agent): reject privileged commands over plain HTTP (#1599)
- 94e6b104d753cb2dc773c5e6e9adba40ba377c84 fix(agent): reject replayed snapshots and serialize updater checks (#1610)
- 89585ba0e2cf1cb048b771f73bd11b324e215e54 fix(agent): require HTTPS for self-heal binary download
- 92edc4a3d302dee7215fefe609d92adb99a15396 fix(agent): run the install one-liner from a downloaded file (#1577)
- d6501960dd699a0ea3939799fc40168ce84955a2 fix(agent): serialize privileged commands (#1593)
- 1724ad2393538b9e89b7e908b9a866774ee409e6 fix(agent): sign RouterOS pusher ingestion (#1571)
- 290d4976406d620b6d73c5c8bc544369c7ba0980 fix(agent): verify AdGuard artifacts and tar contents (#1621)
- 877c654f6d4ecfcc6d4d0994f8611937120367cb fix(alerts): keep the weak-signal 24h episode guard across resolves (#1523)
- 2b4d35626773523e82df34c3c00f1286682495ff fix(auth): enforce API token scopes (#1587)
- d037a8a92292f7bf9b91c56b74db2c94537b9eed fix(auth): restrict router SNMP secrets and alert mutations (#1604)
- 44364a0ce1efa3fe1f1f56e044b51c46586725e1 fix(configbackup): prune snapshots transactionally (#1624)
- 0fafdeae29f612f8ccd7fcb4de938eb6ecf55036 fix(firmware): make upgrade admission atomic (#1594)
- dcd6d8d3ac00b7dec113690f1eb3d1d710bd7b2f fix(firmware): require https and checksum for images (#1616)
- 60c1d27142151c8909ee610de91e90df0450f324 fix(httpapi): bound firmware uploads and agent token bodies (#1606)
- 07bdf089e39cc0f7ca6df8452c00df0f27139ae8 fix(httpapi): lift the deleted-client tombstone on successful known-mac re-add (#1532)
- cdf0eba4c7cee411cd58521dc21d0950f6a3a2e3 fix(orchestr): drop NetGrip apply delegation (#1595)
- 693f40d7adc1fcc0b1a452d7cb0a8ae2d6639959 fix(routerdetail): pick the fastest router-link as a switch/AP backhaul
- d8234347d7c188fa35f1c4a41a4e2b5b4e4a91cc fix(routerdetail): pick the fastest router-link as a switch/AP backhaul (#1555)
- d98240d60730be106dfa4cdb3301afb5efd0d5ec fix(routers): expose real partial vitals (#1569)
- 9573ae507e6818eaa08587f4096078dc55fddbb7 fix(security): secure default bind and PVE TLS verification (#1608)
- 6fe6274251d4b2ac38d37ef9c9f34971e4ebdd6d fix(sse): enforce client limits and swap agent streams atomically (#1612)
- 1ff6615d02f2003d2e8bf726d8712cb2e2af7adf fix(topology): anchor bridge-MAC-less units via registered MAC and LLDP-seeded depths (#1520)
- aa77b4bca1796c7687c926f0a296cbcc76d813cf fix(topology): fill link speeds matching polled ports by ID, Iface and Label (#1512)
- ec2cc1c1de9e8567cdebfe7c42c2de20851ee40c fix(topology): keep a portless wired client on its switch/NAS, not the gateway (#1534)
- c113513b4143b2eb0e33320ef03de863fe20c439 fix(topology): render the WAN-peer chip on PPPoE fleets and stamp wanSide in /api/devices (#1517)
- 8b1b405dc54b717b58af5092effef1a198ccb695 fix(web): avoid duplicate and perpetual channel plan scans (#1567)
- 5118c6561867d35a454e58e7637326be5e369675 fix: bound MQTT shutdown, waitgroups and SSH pool dials (#1614)
- c501f75db83fae94bbcefe9cb7aa3811ab3a58aa refactor(adapters): constants for PeerKind contract values
- 8c0d5fd51eafcf2be2372d009d3e5e136931868d refactor(asuswrt): move support to external integration (#1580)
- 135219f7a85f94330ab8c03684cb54f7ede0e5c2 refactor(routeros): move polling to the external pusher (#1576)
- c39b18d337258626502f5abcaf5d99900dd56ebf revert(devices): drop SNMP switch-port traffic attribution (reverts #1540)
- e6d07f4732545029999b2f004b7ace30234bfe6a revert(routerdetail): drop fastest router-link backhaul picker (reverts #1541)
- d66d6482516a18cfa4c29455d528096b1cc783f3 revert(snmp): drop Netgear FastPath memory + vitals visibility (reverts #1538)
v2.47.0
The portability release, shaped by Bann-DEV's fleet
The agent now works past OpenWrt, and almost everything here exists because Bann-DEV ran NetPulse on an ASUS GT-BE98 Pro (Asuswrt) and a Synology NAS and measured every gap.
The agent goes portable. Without ubus it still reports uptime, RAM (from /proc/meminfo, with the server computing used and percent from the same bytes), timezone (IANA, from /etc/localtime), wireless clients via iw station dump, LLDP neighbours in plain text when the JSON build is missing, CPU temperature via hwmon when a box has no thermal zones, and the WAN port resolved from the default route (LLDP-verified bridge member first). Plus a proper asuswrt router type with the AP role, agent-driven upgrades and no SSH probing. (#1481, #1491, #1498, #1500, #1501)
CPU temperature that means CPU. On x86 the agent reads the package sensor (x86_pkg_temp, coretemp) instead of the first thermal zone, which was usually the ACPI ambient reading. Boxes without zones fall back to hwmon. (#1484)
The agent survives sysupgrades. Both the one-liner installer and the server-side reinstall now pin the binary, env, init script and rc.d symlinks in /etc/sysupgrade.conf — the self-heal needs its entry point to come back. (#1476)
Topology: your bridge ONT, drawn. ARP learned on the gateway's WAN port counts as the WAN peer, so a bridge-mode ONT behind PPPoE appears under Internet instead of vanishing; fleet units without a bridge MAC can be pinned by router id; managed switches stop listing their own CPU MAC as a client; and LLDP links read the uplink speed from the child end when the parent does not report it. (#1485, #1489, #1471)
Presence timeline on Reports. Per-client online spans with a searchable picker, 7/30 day range, online time, span count and connections in 24h — with the open span carried into the window, so always-on hosts no longer plot as empty. (#1493)
Forget SSH host key from the UI. After a reflash or an upgrade that regenerates the keys, clear the pinned key from the unit detail and let the next poll re-register it with trust on first use. (#1477)
Also: a modem/ONT device type (#1495), the releases refresher backs off and surfaces failed checks (#1478), the runtime uplink beats a stale board.json WAN label (#1482), resolved alerts carry their own title and timestamp (#1466), standalone config backup from the unit card (#1462), and both READMEs credit the people behind the project (#1499).
Thanks
Special thanks to Bann-DEV: eight issues (#1481, #1482, #1484, #1485, #1489, #1500, #1501, #1502) and two PRs (#1490, #1491), with measurements on real Asuswrt and Synology hardware and fixes included — the portability work above is largely theirs. And thanks to sweetlazyboy (#1493), crowedavid (the WireGuard, firmware and topology threads) and pl-komuch (#1475); juflo's BBox integration (#1474) is on the way.
Full changelog in CHANGELOG.md. En español:
El agente sale de OpenWrt (#1481, #1491, #1498, #1500, #1501): sin ubus sigue reportando uptime, RAM desde /proc/meminfo (el server calcula usada y porcentaje), zona horaria IANA, clientes WiFi por iw station dump, vecinos LLDP en texto plano, temperatura por hwmon como último recurso y la boca WAN resuelta por ruta default/LLDP. Más el tipo de router asuswrt con rol AP, badge de agente y actualización desde la UI. Temperatura de CPU de verdad (#1484): x86_pkg_temp/coretemp en vez del ambiente ACPI, con hwmon de reserva. El agente sobrevive a sysupgrade (#1476): binario, env, init y symlinks quedan en /etc/sysupgrade.conf al instalar y reinstalar. Topología: tu ONT dibujada (#1485, #1489, #1471): ARP del puerto WAN como peer WAN, unidades fijables por id de router, switches sin su MAC de CPU como cliente y velocidad de uplink desde el extremo hijo. Presencia por cliente en Informes (#1493): buscador, rango 7/30 días y tramos por día. Olvidar clave SSH desde la UI (#1477). Además: tipo Módem/ONT (#1495), refresco de releases con backoff (#1478), uplink runtime manda sobre board.json (#1482), alertas resueltas con su título (#1466), backup standalone (#1462) y READMEs con colaboradores (#1499). Gracias a Bann-DEV (8 issues y 2 PRs, con mediciones en hardware real), sweetlazyboy, crowedavid y pl-komuch; el BBox de juflo está en camino.
v2.46.0
The WireGuard release, shaped by David's fleet
Almost everything in this batch exists because crowedavid runs NetPulse against a real, picky fleet and writes down what he sees.
Peers that matter, alerts that tell the truth. Mark a WireGuard peer as *urgent* and its drops notify like a unit's; the rest just land in the log. Down peers stay visible on the map (dashed ring, card with handshake data), urgent peers get slot priority, and the health score finally cares about VPN state: a dead interface or urgent peers down no longer leave the score at 100, with every penalty explained in the breakdown. (#1441, #1440, #1442)
Connect events you can trust. Tunnels initiated by the local unit no longer lose their connect alert, every reconnection emits its own event (no more 5-minute coalescing hiding the second drop), and the tracker follows the stable public key instead of a dump index that drifted between polls. (#1443, #1445)
Firmware targets that exist. A daily check (configurable) against the official OpenWrt releases index gates the outdated alert: it only fires when the configured target is actually published. A made-up target raises its own configuration warning instead of a permanent alert, and the Firmware page shows the check state with a "check now" button. (#1444)
Topology, tighter. Wired clients behind managed switches are claimed by their switch's FDB instead of dangling off the gateway (#1417); FDB-resolved unit links get their speed from the child end's port table when the parent reports none (#1429); GL.iNet WireGuard peers show their real names and VPN curves anchor cleanly into the Internet node (#1433); Proxmox hosts with zero guests draw as hypervisors (#1446); and the agent (3.0.19) finds LAN bridges that are not br-lan/br0, fixing "0 clients" switches (#1435).
Alerts with a lifecycle. Weak-signal alerts resolve themselves once the client stabilizes above threshold + hysteresis for N consecutive samples (both configurable), keeping duration and recovery value; and clearing or reading alerts updates the overview and badge instantly over SSE instead of waiting for the next poll cycle. (#1452, #1434)
Also: roaming and 802.11r views work on agent-only fleets (#1431, #1450), the NetGrip-embedded units get a proper "Update NetGrip" action (#1436), the landing page is TypeScript now (#1425), and the method selector UI got its review pass: grouped owut steps with tooltips, no leftover Save button, and an expected-SHA256 field for local uploads. (#1414)
Thanks
Special thanks to crowedavid: 23 issues and counting, and reviews so thorough they read like specs - this release is largely his. And thanks to everyone else who filed issues along the way: pl-komuch, TheBox193, Coll147, gnulan and ChaitanyaPramod.
Full changelog in CHANGELOG.md. En español:
Peers que importan y alertas que dicen la verdad (#1441, #1440, #1442): flag de urgencia por peer WireGuard (sus caídas avisan, prioridad en el mapa, penalización en el health score con desglose explicado) y los peers caídos siguen visibles en el mapa. Conecta de verdad (#1443, #1445): los connects de túneles iniciados en local ya no se pierden, cada reconexión emite su evento y el seguimiento va por clave pública estable. Targets que existen (#1444): chequeo diario de releases publicados; la alerta de firmware solo salta si el target existe de verdad, y un target inventado es un aviso de configuración. Topología más fiel (#1417, #1429, #1433, #1436, #1446, #1435): clientes cableados reclamados por su switch, velocidad de enlaces FDB desde el extremo hijo, nombres reales de peers GL.iNet, hosts Proxmox sin guests como hipervisores y agente 3.0.19 con detección de bridge. Alertas con ciclo de vida (#1452, #1434): la señal débil se resuelve sola con histéresis y N muestras, y limpiar/marcar alertas actualiza Resumen al instante por SSE. Además: roaming e itinerancia en flotas agent-only (#1431, #1450), "Actualizar NetGrip" (#1436), landing a TypeScript (#1425) y la revisión del selector de método (#1414). Gracias de nuevo a crowedavid (23 issues y reviews de especificación) y a pl-komuch, TheBox193, Coll147, gnulan y ChaitanyaPramod por sus issues.
v2.45.0
Firmware upgrades, your way
Every unit now picks how it updates: owut (ASU), an image URL, or a local upload of an image you built elsewhere (the server stores it, computes the sha256 and flashes it verified). The owut flow gains explicit steps - *check* to review what would change, *download* to build and fetch the image to the router without flashing, *install* to flash the downloaded one - while the classic "Upgrade now" stays for those who want it in one go. And the config backup no longer dies in the router's /tmp: the agent (3.0.18) uploads it to the server before flashing, the last five per unit are kept, and you can download them right from the unit's card. (#1414)
Alert levels are notification policy now
Setting a category to "Urgent only" used to make everything else vanish from the log. From this release everything is always recorded in the log and the feed; the level only decides what notifies - use the feed's category filters to hide things visually. On top of that, the firmware alert becomes one episode per unit instead of re-firing every 5 minutes, version strings are compared as versions, and a target older than the installed release gets its own possible-downgrade alert. (#1413, #1411, #1412)
Also in this release
- Per-subscriber push language: each push subscription remembers your UI language and the notification is composed in it; subscriptions are user-owned, so unsubscribing only removes your own. (#1334)
- Agent matching with physical criteria: when identical bridged APs share a hostname, the bridge MAC decides; an ambiguous hostname no longer invents an association. (#1334)
- Topology renders all switch-classified routers, not just the first three, and edge units fan their clients toward the center of the canvas. (#1422, #1416)
- SSID impostor: your own fleet no longer reports itself - units whose agent doesn't report own BSSIDs are recognized by their fleet MAC and derived radios. (#1402)
- Firmware manual brought up to date with the actual page. (#1415)
Full changelog in CHANGELOG.md. En español:
Actualizaciones de firmware a tu manera (#1414): método por unidad (owut / URL / subida local), owut por pasos (check → download → install) y copia de configuración en el server (retención 5, descargable desde la tarjeta) en vez de morir en el /tmp del router.
Los niveles de alerta son política de notificación (#1413): todo se guarda siempre en el registro; el nivel solo decide qué notifica. Y la alerta de firmware es un episodio por unidad (#1411), con comparación por versión y aviso de downgrade (#1412).
También: push en tu idioma por suscriptor (#1334), emparejamiento de agentes por MAC física (#1334), todos los switches en la topología (#1422), clientes hacia dentro en los bordes (#1416), la flota ya no se autodenuncia como impostor (#1402) y el manual de firmware al día (#1415).
v2.44.0
Control over the SSID impostor rule
A rule that cries wolf is worse than no rule. From the alert row, the shield button now approves a flagged BSSID as your own: the alert resolves on the spot, the approval persists, and that MAC never trips the rule again (you can revoke it from the approved list). And if you would rather not hear about impostors at all, the alerts settings has a global switch for the whole rule - turning it off also resolves the active impostor alerts. (#1400, #1401)
The rack view graduates out of Labs
The rack gets its own always-visible toggle in Settings > Services (still off by default), the nav item drops the LABS badge and no longer requires Labs, and the caption no longer says "do not use in production". (#1407, #1408)
Also in this release
- #1408: a handful of strings that rendered as raw keys now show real text in both languages: the rack uplink unit selector and its confirmation toast, the close button in Roaming, and the vendor-repository warning in the owut upgrade confirmation.
- #1403, #1405: the public landing folds its feature cards into compact rows with a one-line hook (the details unfold on click) and the install section goes full width.
- #1404, #1406: the English landing screenshots now show the app UI in English, and the demo dataset is advertised as its real 65 devices.
Versión en español
- Regla de SSID impostor domable (#1400, #1401): el botón del escudo en la alerta aprueba el BSSID como propio (se resuelve al momento, queda persistido y no vuelve a saltar), y hay un interruptor global de la regla en los ajustes de alertas: al apagarla se resuelven también las activas.
- La vista de rack sale de Labs (#1407, #1408): toggle propio en Ajustes → Servicios (desactivado por defecto), el nav pierde el badge LABS y la copy suelta el "no usar en producción".
- Además: cadenas que se pintaban como clave cruda ya muestran texto real (selector de unidad de uplink del rack, botón de cerrar en Itinerancia, aviso de vendor en owut), y la landing pública se compacta con capturas EN correctas y el dataset real de 65 dispositivos.
v2.43.0
Two new WiFi security alerts with zero setup: NetPulse now warns when an unknown radio broadcasts one of your SSIDs (evil-twin signature, confirmed across real scans before alerting) and when a brand-new neighbor SSID sticks around for hours - both built on the neighbor scans the fleet already collects.
The activity log keeps its history: resolved alerts no longer disappear - they stay marked with their closing time, and the recovery event expands to show exactly what it resolved.
Also in this release: a new-client alert for devices that join with a known name (the gap "unknown device" could not cover), dragging a switch/AP on the topology map now moves its wired children along, the agent installer configures lldpd on physical ports (bridge-bound lldpd silently sees no neighbors on several OpenWrt targets), links table with per-link ping, a Remove action for dead agents, LLDP evidence hardening (keyed-array parser, root-port anchoring, phantom-neighbor FDB corroboration), and the 802.11r table BSSID fix. Embedded agent 3.0.17 self-upgrades from the server.
Dos alertas WiFi de seguridad sin configuración: NetPulse avisa ahora cuando una radio desconocida emite uno de tus SSIDs (firma de evil twin, confirmada entre scans reales antes de alertar) y cuando un SSID vecino nuevo aguanta horas cerca - ambas sobre los scans de vecinos que la flota ya recoge.
El registro conserva su historia: las alertas resueltas ya no desaparecen - quedan marcadas con su fecha de cierre, y el evento de recuperación se expande mostrando exactamente qué resolvió.
También en esta versión: alerta de cliente nuevo para equipos que entran con nombre conocido (el hueco que "dispositivo desconocido" no cubría), al arrastrar un switch/AP en el mapa sus cableados lo siguen, el instalador del agente configura lldpd en puertos físicos (un lldpd en el bridge no ve vecinos en varios targets OpenWrt y nadie avisaba), tabla de enlaces con ping por enlace, acción Retirar para agentes apagados, endurecimiento de la evidencia LLDP (parser keyed-array, anclaje por puerto raíz, corroboración FDB de vecinos fantasma) y el fix del BSSID en la tabla 802.11r. Agente embebido 3.0.17 auto-actualizable desde el servidor.
v2.42.0
Know when something drops, and why a link exists
Mark any client with the new "Notify when offline" switch in its editor and NetPulse pushes an urgent alert when it actually drops (confirmed over three cycles, self-clearing when it comes back). WireGuard tunnels now log their disconnect too, with the session duration. And every fleet uplink link now declares which evidence produced it (LLDP, FDB, device or manual) instead of labeling everything LLDP - dumb switches behind a managed switch's port appear as inferred nodes, and duplicated switch hover cards collapse.
Also in this release
- Activity log rows now carry a category badge, so the category is readable without guessing from the severity tint (#1357, from #1348 feedback)
- Clients table: the device column gives back space so first/last seen fit on one line; Spanish headers are shorter - and remember you can hide columns from the column picker (#1360)
- Security policy published (SECURITY.md) and monthly Dependabot for dependencies (#1350)
Versión 2.42.0
Saber cuándo algo se cae, y por qué existe un enlace. Marca un cliente con el nuevo check "Avisar si se desconecta" y NetPulse te empuja una alerta urgente cuando se cae de verdad (confirmada en tres ciclos, se retira sola al volver). Los túneles WireGuard también registran su desconexión, con la duración de la sesión. Y cada enlace de unidad de flota declara su procedencia real (LLDP, FDB, device o manual) en vez de etiquetar todo como LLDP; los switches tontos tras un switch gestionado aparecen como nodos inferidos, y las hover cards de switch dejan de duplicarse.
También en esta versión
- Badge de categoría en cada fila del registro de actividad (#1357, feedback de #1348)
- Tabla de clientes: la columna de dispositivo cede espacio y las fechas ya no parten el texto; cabeceras más cortas en español, y recuerda que puedes ocultar columnas desde el selector (#1360)
- Política de seguridad publicada (SECURITY.md) y Dependabot mensual para dependencias (#1350)
v2.41.0
Fleet topology you can trust
A v2.40.0 regression chained fleet units in a loop and left the gateway without links. Fixed: Omada port annotations can no longer act as neighbor identity for uplink resolution (they keep labeling ports), any residual LLDP cycle between units degrades to device/FDB evidence, and semantic link labels now sit on the link midpoint instead of piling up in the corner.
More device icons - additively
Air conditioners, boilers/heating and single-board computers (orange/nano/banana/rock pi) now classify to their own types with dedicated icons (AirVent, Heater, CircuitBoard). Strictly additive: tokens already claimed by an older rule stay put, so no existing device changes its type or icon. Clients also keep their name and icon when a DHCP lease expires, and a MAC is never shown as a hostname anymore.
Also in this release
- Local agent URL behind a reverse proxy:
NETPULSE_AGENT_URLoverrides the public URL in the agent install line, reinstall, re-emit and auto-reinstall (#1335, #1336) - borky's contributions integrated with credit (#1332, #1333): DHCP reservations end-to-end, mDNS host records, finer vendor matching, embedded agent 3.0.15
- Tablet rail: the portrait rail footer now has the expand toggle and the collapsed state works from md up (#1318)
- Expanded detail: the Edit button anchors to the top-right, aligned with the first detail row (#1341)
- New classified device types: clima, caldera, placa (#1340) - requested in discussion #1337
Versión 2.41.0
Topología de flota fiable. Corregida la regresión de v2.40.0 que conectaba las unidades de flota en bucle y dejaba al gateway sin enlaces: las anotaciones de puerto de los Omada ya no identifican vecinos para los uplinks (siguen etiquetando), cualquier ciclo residual degrada a evidencia sólida y las labels de enlaces se colocan en su sitio.
Más iconos de dispositivo, solo añadiendo. Aire acondicionado, caldera y placas SBC (orange/nano/banana/rock pi) tienen tipo e icono propios sin reclasificar nada existente. Los clientes conservan nombre e icono al expirar la concesión DHCP, y una MAC nunca se muestra ya como hostname.
También en esta versión
- URL local para agentes tras reverse proxy:
NETPULSE_AGENT_URL(#1335, #1336) - Integración con crédito del trabajo de borky (#1332, #1333): reservas DHCP, host records mDNS, vendors más finos, agente 3.0.15
- Rail del tablet con toggle y colapsado funcional desde md (#1318)
- Botón Editar del detalle alineado con la primera fila (#1341)
- Nuevos tipos clasificados: clima, caldera, placa (#1340, discusión #1337)
v2.40.0
Topology that listens to what devices announce
The topology graph now trusts what devices announce over the wire before inferring from databases: LLDP outranks FDB evidence when deciding where each fleet unit hangs, resolved with a root-port rule (the neighbor living on the port where the unit learns the gateway's bridge MAC). TP-Link Omada switches - whose remote LLDP table is crippled - now contribute neighbor identity from their local port table (the descriptions you wrote on them), and fleet switches finally show their label on the map.
Manual tagging now actually moves routers, APs and switches: an attach override on a fleet unit's bridge MAC pins its uplink above every inference, accepting router MAC, router id or router:port targets - and uplink links show the negotiated speed and the entry port instead of a generic "1G".
Dedicated icons for CCTV cameras, robot vacuums, and video doorbells
Cameras render the CCTV glyph (and cctv-* hostnames no longer classify as TVs), robot vacuums get their own "aspirador" type out of the generic IoT bucket, and video doorbells get "videoportero" instead of folding into cameras. All three glyphs are selectable in the manual icon picker, and the expanded device detail's Edit button floats off the panel corner.
Also in this release
- #1324 / #1279 - LLDP identity from the local port table (Omada) and LLDP-first uplinks with root-port rule; fleet switch labels on the map
- #1326 / #1321 - manual attach overrides win for fleet units (override > LLDP > device > FDB > gateway), speedMbps on uplink links
- #1328 / #1327 - device types and icons for CCTV cameras, robot vacuums, and video doorbells
- #1330 / #1329 - cctv/vacuum/doorbell glyphs in the manual icon picker; detail Edit button floated off the corner
La topología escucha lo que los equipos se anuncian
El grafo de topología ahora cree lo que los equipos se anuncian por el cable antes que inferirlo de las bases de datos: LLDP manda sobre la evidencia FDB al decidir dónde cuelga cada unidad de la flota, con regla de puerto raíz (el vecino que vive en el puerto donde la unidad aprende la MAC bridge del gateway). Los switches TP-Link Omada - cuya tabla LLDP remota llega capada - aportan ahora la identidad del vecino desde su tabla de puertos local (las descripciones que el admin les puso), y los switches de la flota por fin muestran su etiqueta en el mapa.
El etiquetado manual ahora mueve de verdad routers, APs y switches: un etiquetado sobre la MAC bridge de una unidad de flota fija su uplink por encima de toda inferencia, aceptando destino por MAC del padre, id de router o router:puerto - y los enlaces de uplink muestran la velocidad negociada y el puerto de entrada en vez de un genérico "1G".
Iconos dedicados para cámaras CCTV, aspiradores y videoporteros
Las cámaras pintan el glifo CCTV (y los hostnames cctv-* ya no se clasifican como TVs), los robots aspiradores tienen su propio tipo "aspirador" fuera del cubo IoT genérico, y los timbres con cámara son "videoportero" en vez de mezclarse con las cámaras. Los tres glifos son seleccionables en el picker manual de iconos, y el botón Editar del detalle del dispositivo flota separado de la esquina del panel.
También en esta release
- #1324 / #1279 - identidad LLDP desde la tabla de puertos local (Omada) y uplinks LLDP-first con regla de puerto raíz; etiquetas de switches de flota en el mapa
- #1326 / #1321 - los etiquetados manuales mandan para unidades de flota (manual > LLDP > device > FDB > gateway), speedMbps en los uplinks
- #1328 / #1327 - tipos e iconos para cámaras CCTV, aspiradores y videoporteros
- #1330 / #1329 - glifos cctv/vacuum/doorbell en el picker manual de iconos; botón Editar flotando fuera de la esquina
v2.39.1
Changelog
- b77322101518c23326ca4f49e1ebe31d297a9708 chore: prepare v2.39.1 (#1320)
- 3a611759e0b6b84ef86efa32d250509f9d3b2378 feat(announcements): star campaign with progress bar, button below, accent CTA (#1319)
- 2675733ace5fdd38a53008ebc9228e3aa52b00b6 fix(devices): client lifecycle - lease shown while remaining, hostname/IP retained after lease expiry (#1315) (#1317)
v2.39.0
Changelog
- b686f2bdd10393ce862cb2ac6aad664cb8c84407 chore(deps): bump source-map-js from 1.2.1 to 1.2.2 in /app (#1297)
- 3e05b62770107a69a25c9c6f8aee7e037d440367 chore: prepare v2.39.0 (#1316)
- cf8dc38af0953c9da26b928a9a701b3e3ecd1921 feat(agents): re-emit agent connection strings to the fleet after a TLS change (#1294) (#1313)
- 96a70ee0759fb9f0b0c48e67c0cf813a5b3c35e1 feat(nav): horizontally scrollable mobile bottom nav with all sections (#1314)
- ff3433f8845fe4ad0658b0e9c3fe77399dc57367 fix(deploy): grant CAP_NET_BIND_SERVICE so the TLS listener can bind :443 without root (#1306) (#1311)
- c7db063b8d13b7898916d36c7957671a93ef5a76 fix(devices): purge and filter fleet radio BSSIDs from the client list (#1307) (#1310)
- f8a9dc26b7fb6e3b38ea038bccf6a9a748dddc3d fix(ui): client detail layout, Networks detected sort/filter, sidebar overscroll, survey gateway (#1309)
v2.38.0
English
The Now lens follows the top band selector (#1300): its internal title and band toggle are gone - flip 2.4/5 GHz on the toolbar and the survey follows. The client table sorts band and SSID with a signal-descending tiebreak (strongest neighbors grouped on top), adds a minimum-signal filter (≥ -70/-80/-90 dBm), and linked MACs (#1151) are searchable and visible in the expanded detail. Also restores the sticky band across APs that a working-session merge had partially dropped, and the CI snapshot now builds on every main push (#1271).
Español
La lente "Canales actuales" sigue al selector superior (#1300): su título y toggle de banda internos desaparecen - cambias 2.4/5 GHz en la barra y el survey la sigue. La tabla de clientes ordena banda y SSID desempatando por señal descendente (los vecinos fuertes agrupados arriba), añade filtro de señal mínima (≥ -70/-80/-90 dBm), y las MACs enlazadas (#1151) son buscables y visibles en el detalle. Además restaura la banda sticky entre APs que un merge de sesión concurrente dejó a medias, y el snapshot de CI se compila en cada push a main (#1271).
v2.37.0
English
Real WAN latency on OpenWrt (agent 3.0.14): the ping parser only understood GNU iputils' 4-number summary; busybox prints 3 - every OpenWrt install showed an eternal em dash in the latency cards. Now it measures.
Also in this release: the Channels unit picker drops units with no WiFi bands; the picked band stays when you switch APs; the lens toggle reads "Analysis / Current channels"; the sidebar shows version · build, adds Fleet/Clients attention badges and scrolls on short portrait screens; SNMP ports show 2.5 Gbps instead of a truncated "2 Gbps"; turning HTTPS on with a custom port applies the port before binding (no more ":443 permission denied" on non-root installs); deleted clients stay deleted (tombstones filter the final list too), fleet units' own MACs are purged from the client registry, and deleting a router cleans up its agent stub and revokes the token.
Español
Latencia WAN real en OpenWrt (agente 3.0.14): el parser del ping solo entendía el resumen de 4 números de GNU iputils; busybox imprime 3 - toda instalación OpenWrt mostraba una raya eterna en las tarjetas de latencia. Ahora mide.
Además: el selector de Canales excluye unidades sin bandas WiFi; la banda elegida se mantiene al cambiar de AP; el selector de lente dice "Análisis / Canales actuales"; el menú muestra versión · build con badges de atención en Flota y Clientes y scroll en pantallas cortas; SNMP muestra 2.5 Gbps sin truncar; activar HTTPS con puerto personalizado aplica el puerto antes del bind (adiós al ":443 permission denied" en no-root); los clientes borrados se quedan borrados (lápidas también en la lista final), las MACs de la propia flota se purgan del registro y borrar un router limpia su stub y revoca el token.
v2.36.0
English
The Now lens works on agent-only fleets (#1270, agent 3.0.13): the channel survey used to run over SSH, so units without SSH credentials (agent-first installs) never had data and the lens was born empty. The agent now runs iw dev <iface> survey dump itself during its full probe (5-minute throttle, forced by the refresh button) and pushes the raw text; the server caches it for 15 minutes and the lens merges it, reusing the existing parser. SSH-based units keep their current path.
Español
La lente "Ahora" funciona en flotas solo-agente (#1270, agente 3.0.13): el survey de ocupación iba por SSH, así que las unidades sin credenciales SSH (instalaciones solo-agente) no tenían datos jamás y la lente nacía vacía. Ahora el propio agente ejecuta iw dev <iface> survey dump en su sondeo completo (throttle de 5 minutos, forzado por el botón Actualizar) y empuja el texto crudo; el servidor lo cachea 15 minutos y la lente lo fusiona reutilizando el parser existente. Las unidades con SSH siguen su camino actual.
v2.35.0
English
Channels analysis rebuilt (#1214): LuCI-style spectrum with a gentle curve, adaptive dBm axis with real gridlines, per-SSID colors for every network (own included), hover shading that reaches networks nested inside bigger ones, occupancy strip with real bandwidth overlap, persistent table sort, and a "Scan now" button with a 30-day fallback for fleets that don't scan periodically.
Also: the service worker no longer serves a stale app shell - one normal reload is always enough after an update; a build stamp in the sidebar tells exactly which commit your browser runs; SNMP gets full device names and MikroTik SwOS chassis temperature; per-radio noise and BSSID telemetry (agent 3.0.11); conservative channel suggestions that only move you when the gain is clear.
Fixes worth mentioning: the spectrum no longer pegs every network at the top; the updater no longer reports phantom updates on rolling installs; CSV export and offline-only delete in Clients; WAN latency is honest (target configurable, "—" when unknown).
Thanks @crowedavid for the reports that fed this batch.
Español
Análisis de canales rehecho (#1214): espectro estilo LuCI con curva, eje dBm adaptativo con rejillas reales, colores por SSID para todas las redes (propias incluidas), sombreado al pasar el ratón que llega también a redes anidadas dentro de otras, franja de ocupación con solapamiento real, orden persistente en la tabla y botón "Escanear ahora" con fallback de 30 días para flotas sin scan periódico.
Además: el service worker ya no sirve la app vieja - con un reload normal basta tras actualizar; sello de build en el menú para saber qué commit ejecuta tu navegador; SNMP con nombre completo y temperatura de chasis MikroTik SwOS; telemetría de ruido y BSSID por radio (agente 3.0.11); sugerencias de canal conservadoras que solo te mueven cuando la ganancia es clara.
Arreglos destacados: el espectro ya no pega todas las redes al techo; el updater no anuncia actualizaciones fantasma en modo rolling; export CSV y borrado solo-offline en Clientes; la latencia WAN es honesta (objetivo configurable, "—" si no hay dato).
Gracias @crowedavid por los reportes que alimentaron esta tanda.
v2.34.0
v2.34.0 - Guided help, a usable updater and honest interface names
- Guided integration flows in the Help page (#1184): step-by-step in-app walkthroughs for MQTT/Home Assistant, ntfy, Telegram and the MCP server, in English and Spanish.
- The "Update NetPulse" dialog works again on short viewports (#1197): the action buttons used to end up off-screen with no way to reach them. The dialog now caps at the viewport at every size and its body scrolls. Reported by @crowedavid.
- Real interface names on OpenWrt (#1200): the BPI-R4's
sfp-lan/sfp-wan(netdev-name, OpenWrt 25.12+) showed as "sfp-LAN "/"SFP -wan", and the UniFi 6 Lite's barelanport left the ports panel empty (0 of 0). Classic names keep their friendly form (LAN 1, WAN, ETH 0, SFP 0); anything else shows its real interface name. Agent 3.0.10. Reported by @pl-komuch - thanks for the screenshots, they made the diagnosis trivial.
The CHANGELOG.md now covers every release again (it had fallen behind at 2.31.1).
Upgrade: curl -fsSL https://raw.githubusercontent.com/gnacho/netpulse/main/install.sh | bash - or update from Settings.
v2.34.0 - Ayuda guiada, un updater usable y nombres de interfaz honestos
- Flujos guiados de integraciones en la Ayuda (#1184): recorridos paso a paso dentro de la app para MQTT/Home Assistant, ntfy, Telegram y el servidor MCP, en castellano y en inglés.
- El diálogo "Actualizar NetPulse" vuelve a funcionar en pantallas cortas (#1197): los botones se quedaban fuera de pantalla sin forma de llegar; ahora el diálogo se ajusta al viewport en todos los tamaños y el cuerpo recorre. Reportado por @crowedavid.
- Nombres de interfaz reales en OpenWrt (#1200):
sfp-lan/sfp-wandel BPI-R4 se mostraban como "sfp-LAN "/"SFP -wan", y la bocalana secas del UniFi 6 Lite dejaba el panel de puertos vacío (0 de 0). Los nombres clásicos conservan su formato (LAN 1, WAN, ETH 0, SFP 0); el resto muestra su nombre real. Agente 3.0.10. Reportado por @pl-komuch: gracias por las capturas, dejaron el diagnóstico trivia.
El CHANGELOG.md vuelve a cubrir todas las versiones (se había quedado en 2.31.1).
Actualizar: curl -fsSL https://raw.githubusercontent.com/gnacho/netpulse/main/install.sh | bash - o desde Ajustes.
v2.33.0
v2.33.0 — Clients that never get lost, links that show
⭐ The main event: clients first
Three changes that shift the day-to-day:
- Indefinite client retention: discovered devices no longer disappear from the list when their lease expires or the ARP table gets cleared. They stay listed as offline with their last known name, IP and last seen (#1145).
- Delete client: a new action in the edit sheet, with confirmation. A deleted client only comes back if seen on the network again.
- One client, multiple MACs: devices that use a different MAC per network (a watch, a laptop) can now be linked into a single row (#1151).
Also
- Fleet unit pause: pause a unit without removing it — it stops being polled, stays quiet and wears its own badge (#1085).
- Topology and rack: uplinks now derive from the managed switch's FDB (no more flat star or invented cables); rack mounts show the real fleet names; unidentified devices are no longer offered as candidates; and a new uplink unit designation for the auto-cabling, which syncs itself when the physical cabling changes. This feature was designed with the community in discussions #1108 (draw your physical rack) and #1109 (auto-cabling from detected topology).
- Real latency: the gateway latency card shows the measured 24 h series (before: a fake 0 ms with a demo chart) (#1148).
- Docs: the landing and READMEs cover channel analysis, MCP and the rack (#1152).
Fixes
- MQTT: the publisher uses gonzalop/mq v0.9.11 (goodbye to the reconnect storm that froze Home Assistant for hours), waits for the old publisher to shut down on hot reconfig, and guards its internal state with a mutex (#1153, #1158, #1162).
- Ports and LLDP: port tooltips no longer clip IP/MAC/detail and show the LLDP neighbor info (#1149, #1150).
- Channels: the picker only offers APs (#1147, partial: the ad hoc scan and conservative suggestions remain).
- Firmware upgrades: out of Labs, with its own visibility toggle (#1146).
- Security and compliance: THIRD_PARTY.md with dependency licenses included in the release archives, golang.org/x/crypto bump (2 SSH CVEs) and other dependency updates, race fixes in the RTL console scraping and in the tests (#1167–#1173).
Special thanks to @crowedavid: the client retention, linking, LLDP and latency work in this release started with his reports.
v2.33.0 — Clientes que no se pierden, enlaces que se ven
⭐ Lo principal: los clientes mandan
Tres mejoras que cambian el día a día:
- Retención indefinida de clientes: los equipos descubiertos ya no desaparecen de la lista cuando caduca su lease o se limpia la tabla ARP. Permanecen como offline con su último nombre, IP y última vez vistos (#1145).
- Borrar cliente: nueva acción en la hoja de edición, con confirmación. Un cliente borrado solo reaparece si se le vuelve a ver en la red.
- Un cliente, varias MAC: los dispositivos que usan una MAC distinta por red (un reloj, un portátil) ahora se enlazan en una sola fila (#1151).
Además
- Pausa de unidades de flota: pausa una unidad sin borrarla — deja de sondearse, no alerta y se distingue con su insignia (#1085).
- Topología y rack: los uplinks salen ahora del FDB del switch gestionado (adiós a la estrella plana y a los cables inventados); los montajes del rack muestran los nombres reales de la flota; los equipos no identificados ya no se ofrecen como candidatos; y nueva designación de unidad central para el auto-cableado, que se sincroniza solo cuando cambia el cableado físico. Esta función se diseñó con la comunidad en las discusiones #1108 (draw your physical rack) y #1109 (auto-cabling from detected topology).
- Latencia real: la tarjeta de latencia del gateway muestra la serie de 24 h medida (antes: un 0 ms falso con gráfica de demo) (#1148).
- Documentación: landing y READMEs al día (análisis de canales, MCP, rack) (#1152).
Correcciones
- MQTT: el publisher usa gonzalop/mq v0.9.11 (adiós a la tormenta de reconexiones que congelaba Home Assistant durante horas), espera el apagado del publisher viejo al reconfigurar en caliente y protege su estado interno con mutex (#1153, #1158, #1162).
- Puertos y LLDP: los tooltips de los puertos ya no recortan IP/MAC/detalle y muestran la información del vecino LLDP (#1149, #1150).
- Canales: el desplegable solo ofrece APs (#1147, parcial: quedan el scan ad hoc y las sugerencias conservadoras).
- Firmware upgrades: fuera de Labs, con su toggle propio de visibilidad (#1146).
- Seguridad y cumplimiento: THIRD_PARTY.md con las licencias de dependencias incluido en los archivos de release, bump de golang.org/x/crypto (2 CVEs SSH) y otras dependencias, fixes de carreras en el sondeo RTL y en los tests (#1167–#1173).
Gracias a @crowedavid: el trabajo de retención, enlazado, LLDP y latencia de esta release empezó con sus reportes.
All PRs in this release / Todos los PRs de esta release
#1143 · #1154 · #1155 · #1156 · #1160 · #1161 · #1163 · #1164 · #1165 · #1166 · #1168 · #1171 · #1172 · #1173 · #1174 · #1176 · #1179 · #1187 · #1188 · #1189 · #1190 · #1191 · #1192 · #1193 · #1194 · #1195
Instalación / Upgrade: curl -fsSL https://raw.githubusercontent.com/gnacho/netpulse/main/install.sh | bash — o actualiza desde Ajustes / or update from Settings.
v2.32.0
Highlights
1. Embedded MCP server (labs) — your network, from your AI assistant
NetPulse now ships with a built-in read-only MCP server at /mcp: fleet
status, devices, alerts, topology and more, queried by your AI assistant
over HTTP with an API token. Labs-only, off by default, rate-limited and
audited.
2. LLDP-MIB topology on SNMP switches
Managed SNMP switches now use LLDP-MIB for topology discovery — proper
uplink visibility on hardware that never exposed an OpenWrt bridge.
This one exists thanks to @crowedavid — literally, without him and his
hardware this would not have been possible.
3. Rack canvas (experimental preview, labs)
A physical rack view: draw your 19" rack, place your routers, switches
and servers, and let NetPulse detect the cabling from the FDB —
including router-to-router uplinks. Very early preview: disabled by
default (enable it in Settings → Labs), still under heavy development.
Do not use it in production.
Also in this release
- Firmware Upgrades now lists only vanilla OpenWrt devices, with an
explanatory note — and graduates out of labs (#1121)
- Alerts: the unread-only view option now persists (#1122), and the
read mark is forgotten when an alert is resolved (#1133)
- Speedtest: adhoc button and "testing" state detection restored (#1132)
- Roaming: events client filter matches the details view and filters
are now multi-select (#1139)
- UI: AdGuard and AP markers no longer shown where they do not apply (#1141)
- SNMP: virtual interfaces grouped instead of hidden (#1142, closes #1125)
- SNMP: whitespace-only ifAlias port labels ignored (#1077)
- Overview: first/last seen applied to the SSE device snapshot (#1111)
- Update dialog resized to match Pulse, with blurred backdrop (#1116)
Lo más destacado
1. Servidor MCP integrado (labs) — tu red, desde tu asistente IA
NetPulse incorpora un servidor MCP de solo lectura en /mcp: estado de
la flota, dispositivos, alertas, topología y más, consultables por tu
asistente IA por HTTP con token de API. Solo en Labs, apagado por
defecto, con rate-limit y auditoría.
2. Topología LLDP-MIB en switches SNMP
Los switches gestionados por SNMP ya usan LLDP-MIB para el descubrimiento
de topología: enlaces reales en hardware que nunca expuso un bridge
OpenWrt. Esta funcionalidad existe gracias a @crowedavid — literalmente,
sin él y su hardware esto no habría sido posible.
3. Rack canvas (vista previa experimental, labs)
Una vista física de tu rack: dibuja tu mueble de 19", coloca routers,
switches y servidores, y deja que NetPulse detecte el cableado desde el
FDB — uplinks entre routers incluidos. Vista previa muy temprana:
desactivada por defecto (actívala en Ajustes → Labs), aún en pleno
desarrollo. No la uses en producción.
Además en esta versión
- Actualizaciones muestra solo dispositivos con OpenWrt vanilla, con
nota explicativa — y sale de labs (#1121)
- Alertas: la opción "solo no leídas" ahora persiste (#1122) y se olvida
la marca de leída al resolverse una alerta (#1133)
- Speedtest: botón adhoc y detección de "testing" restaurados (#1132)
- Roaming: el filtro de cliente de los eventos coincide con el detalle
y los filtros son multiselección (#1139)
- Interfaz: los marcadores AdGuard y AP ya no se muestran donde no
aplican (#1141)
- SNMP: las interfaces virtuales se agrupan en vez de ocultarse (#1142,
cierra #1125)
- SNMP: se ignoran ifAlias de solo espacios (#1077)
- Resumen: first/last seen aplicado al snapshot SSE (#1111)
- Diálogo de actualización con tamaño y fondo de Pulse (#1116)
Full changelog / Changelog completo
- #1138 — fix(rack): cables render via DOM-measured overlay
- #1139 — feat(rack): labs gating + fix(roaming): events filter multi-select
- #1135 — feat(rack): land the rack canvas iteration on main
- #1133 — fix(alerts): forget the read mark on resolve
- #1132 — fix(speedtest): adhoc button and testing detection
- #1130 — fix(rack): resolve conflict markers in Faceplate
- #1128 — fix(rack): fleet uplink hints from raw FDB evidence
- #1142 — feat(snmp): group virtual interfaces instead of hiding them
- #1141 — fix(ui): stop showing AdGuard and AP markers where they do not apply
- #1123 — feat(rack): physical rack canvas with auto-cabling (labs)
- #1122 — feat(alerts): persist the unread-only view option
- #1121 — feat(firmware-upgrades): only list vanilla OpenWrt devices
- #1120 — feat(settings): mark the MCP integration as labs-only
- #1119 — chore: ignore locally built binaries
- #1118 — feat(mcp): embedded read-only MCP server at /mcp
- #1117 — feat(snmp): use LLDP MIB on SNMP switches for topology discovery
- #1116 — fix(home): size the update dialog like Pulse and blur the backdrop
- #1111 — fix(overview): apply first/last seen to the SSE device snapshot
- #1077 — fix(snmp): ignore whitespace-only ifAlias port labels
v2.31.1
English
Patch release with the first community feedback on v2.31.0:
- Alerts that re-fire right after a server restart start unread again, instead of inheriting the read state from the previous session (#1094).
- First seen / Last seen no longer show a permanent dash; the seen marks are now written through the same device list the UI consumes (#1095).
- Roaming events from the same second keep their log order (OpenWrt logs have no milliseconds; ties break by insertion order), and the events view gains client and AP dropdowns to follow one client across the fleet (#1097).
- The client edit sheet keeps Save/Cancel next to the edited fields, and the card is titled "Edit client" (#1096).
- The unit's Open web UI and SSH copy buttons show the exact URL/command in a tooltip before you press them (#1098).
- Agent 3.0.8: a unit with no clients associated at push time no longer drops its whole wireless section (radios, scans, own BSSIDs).
Español
Release de parche con los primeros comentarios de la comunidad sobre la v2.31.0:
- Las alertas que se re-emiten justo tras un reinicio del servidor vuelven a aparecer como no leídas, en vez de heredar el estado de la sesión anterior (#1094).
- "First seen" / "Last seen" ya no muestran un guion permanente; las marcas de visto ahora se escriben desde la misma lista de dispositivos que consume la UI (#1095).
- Los eventos de roaming del mismo segundo conservan el orden del log (los logs de OpenWrt no llevan millis; el desempate es el orden de inserción), y la vista de eventos gana desplegables de cliente y AP para seguir a un cliente entre APs (#1097).
- La tarjeta de edición de cliente mantiene Guardar/Cancelar junto a los campos que editan, y el título pasa a "Editar cliente" (#1096).
- Los botones de web y SSH copiar de la unidad muestran la URL/comando exacto en un tooltip antes de pulsar (#1098).
- Agente 3.0.8: una unidad sin clientes asociados en ese momento ya no descarta toda su sección wireless (radios, scans, BSSIDs propios).
v2.31.0
English
Highlight 1: Channels graduates from labs - a read-only WiFi spectrum analyzer
The channel analysis leaves labs and becomes a first-class page. It never writes to your routers: it reads the air and advises, you decide on the router.
- A per-radio waterfall: every network drawn as a bell on a real frequency axis, tinted DFS ranges, your own networks highlighted at their real channel width, a marker on the suggested channel, and a continuous occupancy heat strip.
- A score for every channel: the strip below the chart groups all 2.4 GHz channels and every 5 GHz block (DFS included, tagged) by congestion - optimal, moderate, busy. Click one to focus it on the spectrum.
- A smarter congestion engine: your own networks never congest a channel (detected via MAC prefixes, SSID transitivity, dawn/usteer data and the new agent-reported own BSSIDs), and neighbors are weighted in the power domain, so one loud AP outweighs four quiet toys. The verdict for your current channel is absolute, not relative to the rest of the band.
- A colorful, sortable network table with signal bars, own-network badges and click-to-isolate on the chart.
- Suggestion only: the page recommends a channel and estimates the noise reduction; applying it stays in your hands, on the router.
No labs badge, no toggle: it is in the sidebar for everyone, ready to use.
Highlight 2: GitHub Discussions is open
We have opened GitHub Discussions for the project. Suggest features, ask questions, show your setup, and vote on ideas - the backlog listens, and popular requests get priority: https://github.com/gnacho/netpulse/discussions
Also in this release
- Fixed the "Agent outdated" push notification showing a static 16:00:00 timestamp (epoch in the server timezone) when it fired right after a server restart (#1074).
- Fixed the channel analysis briefly reporting "this router has no WiFi radios" when the agent payload was momentarily stale.
- Agent 3.0.7: reports its own BSSIDs and reads the advertised channel width (HT/VHT) of every neighbor, so the spectrum draws each network at its real width.
Español
Lo destacado 1: Canales sale de labs - un analizador de espectro WiFi de solo lectura
El análisis de canales deja labs y se convierte en una página de primera clase. Nunca escribe en tus routers: lee el aire y aconseja; la decisión sigue siendo tuya, en el router.
- Cascada por radio: cada red se dibuja como una campana sobre un eje de frecuencia real, con zonas DFS sombreadas, tus propias redes resaltadas a su ancho de canal, una marca en el canal sugerido y una franja continua de ocupación.
- Puntuación de cada canal: la tira bajo el gráfico agrupa los 13 canales de 2.4 GHz y todos los bloques de 5 GHz (DFS incluidos, etiquetados) por congestión - óptimos, moderados, saturados. Clic para enfocarlos en el espectro.
- Un motor de congestión más listo: tus redes propias nunca congestian un canal (detectadas por prefijo MAC, transitividad de SSID, datos dawn/usteer y los nuevos BSSIDs propios que reporta el agente), y las vecinas se ponderan en dominio de potencia: un AP fuerte pesa más que cuatro de juguete. El veredicto de tu canal actual es absoluto, no relativo al resto de la banda.
- Tabla de redes ordenable, con barras de señal, insignias de redes propias y clic para aislar en el gráfico.
- Solo sugiere: la página recomienda canal y estima la reducción de ruido; aplicarlo sigue en tus manos, en el router.
Sin insignia de labs ni toggle: está en el menú para todo el mundo, lista para usar.
Lo destacado 2: GitHub Discussions está abierto
Hemos abierto GitHub Discussions para el proyecto. Sugiere funciones, pregunta, enseña tu instalación y vota ideas - el backlog escucha, y las propuestas con más apoyo tienen prioridad: https://github.com/gnacho/netpulse/discussions
También en esta versión
- Corregida la notificación push de "Agent outdated" con la hora estática 16:00:00 (epoch en la zona del servidor) al dispararse justo tras un reinicio (#1074).
- Corregido el análisis de canales mostrando "este equipo no reporta radios WiFi" cuando el payload del agente estaba momentáneamente desactualizado.
- Agente 3.0.7: reporta sus propios BSSIDs y lee el ancho de canal anunciado (HT/VHT) de cada vecina, así el espectro dibuja cada red a su ancho real.
v2.30.0
English
RouterOS (MikroTik) support, first class (#969, thanks @samex). New fleet router type polled server-side via the native REST API of RouterOS 7+ (real CPU/RAM/uptime vitals, WAN metrics, package update check), with a hand-rolled binary API client (port 8728) for installs that keep the web interface closed: zero new dependencies, modern login, read-only sentences. The password stays server-side and is never serialized back, factory self-signed TLS is opt-in, and an optional routeros-pusher agent covers api-only units.
M-Lab NDT is now the default speedtest (#1037, suggested by @crowedavid). A neutral, non-profit measurement network with an open data policy: a better privacy story than Ookla or Cloudflare. Installs with a saved provider keep it, and a new "auto" schedule runs the test at the quietest hour according to your own traffic history (#1066).
Topology:
- Fleet uplinks anchor to their real parent from switch FDB evidence, with the port label (#1047, #1051).
- When the evidence port has an inferred or managed switch circle, the uplink hangs from the circle: gateway to (switch) to AP (#1060).
- The WAN upstream device (ISP modem/ONT) is drawn under the Internet node when its IP matches the WAN gateway (#1042).
- LLDP matching tolerates hostname vs FQDN (#1041); the Proxmox integration absorbs spurious inferred switches (#1053).
- No more hover previews while editing the layout (#1045); the links table agrees with the map (#1063).
Fixes from your reports (thanks @crowedavid and @gnulan):
- Roaming feed shows Connect events on wpad-mbedtls APs (#1038).
- Blank screen when switching report ranges fixed (#1029); hourly and daily granularity (#1032, #1033).
- Fleet edit card: changing the type no longer saves and closes; visible labels; firmware target only for OpenWrt/GL.iNet (#1026).
- "Mark all as read" no longer clears the log: separate clear-log button with confirmation and a retention setting (#1034).
- Proxmox: duplicate or empty instance ids are refused (#1057); the detect button is back (#967).
- Traffic history port selector becomes a dropdown with an All/Active/Inactive filter (#1035).
- Port cards: no more hardcoded Spanish strings leaking into other languages; guessed peers carry a network icon (#1039, #1040).
- Alerts are composed in English natively and translated to your language (#1014); release notes show in the client language (#1011).
- Full dependency refresh (#1020, #1022, #1028).
- Dialog polish: close on save, hints collapsed, no uninvited tooltips, focus ring off the close button and on hover (#1064 to #1066, #1068, #1069).
Huge thanks to @crowedavid, whose sharp issue reports shaped most of this release; to @samex for the RouterOS integration; and to @gnulan, @borky and everyone else reporting bugs and discussing ideas. This is what open source is about.
Español
Soporte RouterOS (MikroTik) de primera clase (#969, gracias @samex). Nuevo tipo de router en la flota, sondeado desde el servidor por la API REST nativa de RouterOS 7+ (vitals reales de CPU/RAM/uptime, métricas WAN, comprobación de actualizaciones), con cliente del API binario (puerto 8728) escrito a mano para instalaciones con la interfaz web cerrada: cero dependencias nuevas, login moderno y sentencias de solo lectura. La contraseña solo vive en el servidor y nunca se serializa de vuelta, el TLS autofirmado de fábrica es opt-in, y un agente routeros-pusher opcional cubre unidades solo-api.
M-Lab NDT es ahora el speedtest por defecto (#1037, idea de @crowedavid). Una red de medición neutra y sin ánimo de lucro, con política de datos abierta: mejor privacidad que Ookla o Cloudflare. Las instalaciones con proveedor guardado lo conservan, y un nuevo modo "auto" ejecuta el test a la hora de menos tráfico según el histórico propio de la red (#1066).
Topología:
- Los uplinks de la flota se anclan a su padre real según el FDB de los switches, con etiqueta de puerto (#1047, #1051).
- Si el puerto de la evidencia tiene un círculo de switch (inferido o gestionado), el uplink cuelga del círculo: gateway a (switch) a AP (#1060).
- El equipo aguas arriba del gateway (módem/ONT del ISP) se dibuja bajo el nodo Internet cuando su IP coincide con la puerta de enlace WAN (#1042).
- El matching LLDP tolera hostname vs FQDN (#1041); la integración Proxmox absorbe switches inferidos espurios (#1053).
- Sin previsualizaciones al editar el layout (#1045); la tabla de enlaces coincide con el mapa (#1063).
Arreglos de vuestros reportes (gracias @crowedavid y @gnulan):
- El feed de roaming muestra eventos Connect en APs con wpad-mbedtls (#1038).
- Arreglada la pantalla en blanco al cambiar de rango (#1029); granularidad horaria y diaria (#1032, #1033).
- Tarjeta de edición de flota: cambiar el tipo ya no guarda y cierra sola; etiquetas visibles; firmware target solo para OpenWrt/GL.iNet (#1026).
- "Marcar todo como leído" ya no vacía el registro: botón aparte de vaciar registro con confirmación y retención configurable (#1034).
- Proxmox: se rechazan ids de instancia vacíos o duplicados (#1057); el botón detectar está de vuelta (#967).
- El selector de puertos del historial de tráfico es ahora un desplegable con filtro All/Active/Inactive (#1035).
- Tarjetas de puertos: se acabaron los textos en español filtrándose a otros idiomas; los pares inferidos llevan icono de red (#1039, #1040).
- Las alertas se componen en inglés y se traducen a tu idioma (#1014); las notas de release se muestran en el idioma del cliente (#1011).
- Actualización completa de dependencias (#1020, #1022, #1028).
- Pulido de diálogos: cerrar al guardar, hints plegados, tooltips que no saltan solos, foco fuera de la (x) y anillo al pasar el ratón (#1064 a #1066, #1068, #1069).
Enormes gracias a @crowedavid, cuyos reportes afilados formaron la mayor parte de esta release; a @samex por la integración RouterOS; y a @gnulan, @borky y todos los que reportan bugs y discuten ideas. De esto va el open source.
v2.28.43
English
This one is big. It started as a layout tweak and turned into a full UX pass over the whole app, driven by feedback from people actually running it. Agent unchanged (3.0.5).
The overview has a new layout (#965, #979): a compact hero with the greeting, latency and clients on one side and the health ring on the other, alerts next to it with icon-only actions, and WAN traffic at full width with the measured-speed graph back in, plus the last test line: plan %, down/up, ping, age, server (#982). The routers section is now "Your fleet", AdGuard shows before WireGuard, and an empty alerts card finally looks like something (#992).
Settings has a new Services and integrations card (#964, #968): AdGuard, WireGuard and OpenVPN as network services, MQTT and Proxmox as integrations, all with server-side toggles and proper configure dialogs. ntfy and Telegram moved to Notifications, where they belong (#996). There's AdGuard autodetection (#964), Proxmox subnet discovery in two phases because pveproxy rate-limits parallel TLS (#967), a configurable HTTPS port that rebinds live, default 443 (#978), and you can finally pick the speedtest provider: Ookla, Cloudflare, LibreSpeed or your own endpoint (#976, #1001).
Alerts behave better: "mark all as read" actually empties the feed (#971), and agent-down alerts are volatile per router instead of piling up in the log (#966). Reports are honest again: the availability math assumed a 5s poll and reported up to 17 days of phantom downtime on perfectly healthy routers, so it now uses real 5-minute bucket coverage (#987). The page explains what it measures, shows downtime in hours and minutes, and has an uptime-kuma style bar with date and time tooltips (#973, #993).
Clients: fleet and connection filters are multi-select dropdowns that scale past a handful of routers (#970, #989), you can hide table columns, name excepted, and the freed space is shared evenly (#958). 6 GHz only shows up if your network actually has it (#991). Device detail lost the dead weight (#985).
The whole settings page went on a text diet (#1002): the long lectures moved into (i) tooltips or the Help page, the HTTPS card is a toggle and a port (#995), trusted devices and API tokens are just add + list (#999, #1000), and the update history lives under About > System. The update dialog itself looks like Pulse's now. Labs toggles are red because they write to your routers.
A lot of this release exists because @crowedavid keeps filing careful, reproducible reports, this time the weak-signal threshold (#1003). His steady feedback over the last weeks has shaped a good chunk of the app, and it shows. Thanks, David. Thanks also to @gnulan for the router-name report (#988).
Español
Esta es grande. Empezó como un retoque de layout y acabó siendo una pasada de UX completa a toda la app, guiada por feedback de gente que la usa de verdad. El agente no cambia (3.0.5).
El resumen estrena layout (#965, #979): hero compacto con el saludo, latencia y clientes a un lado y el anillo de salud al otro, las alertas al lado con acciones solo icono, y el tráfico WAN a todo lo ancho con la gráfica de velocidad medida de vuelta, más la línea del último test: % del plan, bajada/subida, ping, antigüedad y servidor (#982). La sección de routers ahora es "Tu flota", AdGuard aparece antes que WireGuard y la tarjeta de alertas vacía por fin tiene buen aspecto (#992).
Ajustes tiene una tarjeta nueva de Servicios e Integraciones (#964, #968): AdGuard, WireGuard y OpenVPN como servicios de red, MQTT y Proxmox como integraciones, todos con toggles en el servidor y diálogos de configuración como dios manda. ntfy y Telegram se mudan a Notificaciones, que es su sitio (#996). Hay detección de AdGuard (#964), descubrimiento de Proxmox en la subred en dos fases porque pveproxy limita el TLS paralelo (#967), puerto HTTPS configurable que se rebinda en caliente, por defecto 443 (#978), y por fin puedes elegir el proveedor del test de velocidad: Ookla, Cloudflare, LibreSpeed o tu propio endpoint (#976, #1001).
Las alertas se comportan mejor: "marcar todo como leído" vacía el feed de verdad (#971) y las de agente caído son volátiles por router en vez de acumularse en el log (#966). Los informes vuelven a ser honestos: la disponibilidad asumía un sondeo de 5 segundos y llegó a reportar 17 días de caída fantasma en routers sanos; ahora usa la cobertura real de buckets de 5 minutos (#987). La página explica qué mide, muestra el tiempo sin datos en horas y minutos y tiene una barra estilo Uptime Kuma con tooltips de fecha y hora (#973, #993).
Clientes: los filtros de flota y conexión son desplegables multi-selección que escalan más allá de un puñado de routers (#970, #989), puedes ocultar columnas de la tabla, menos la del nombre, y el espacio liberado se reparte a partes iguales (#958). La banda 6 GHz solo aparece si tu red la tiene de verdad (#991). El detalle del dispositivo perdió lastre (#985).
Toda la página de Ajustes se puso a dieta de textos (#1002): los discursos largos se mudaron a tooltips de información o a la Ayuda, la tarjeta HTTPS es un toggle y un puerto (#995), dispositivos de confianza y tokens de API son añadir + listado (#999, #1000), y el historial de actualizaciones vive en Acerca de > Sistema. El diálogo de actualización ahora es estilo Pulse. Los toggles de Labs van en rojo porque escriben en tus routers.
Gran parte de esta release existe porque @crowedavid no para de enviar reportes cuidadosos y reproducibles, esta vez el umbral de señal débil (#1003). Su feedback constante de las últimas semanas ha dado forma a buena parte de la app, y se nota. Gracias, David. Gracias también a @gnulan por el reporte de los nombres de router (#988).
v2.28.42
What's Changed
Fixed
- Apply button blocked on installs without a git clone (#955) - the "clean working tree" readiness check ran on any rolling-mode install (detected by the presence of
deploy/update.sh) even when the directory was not a git clone; it failed with "could not read git status" and blocked the apply button forever, even thoughupdate.shfully supports that layout. The check is now skipped when there is no.git, same as the stable layout. - Connectivity probe exhausted the anonymous GitHub API quota (#956) - the readiness network check called api.github.com on every compute (cached only 30 s); with the update UI open it could exceed the 60 req/h anonymous quota, and the apply then failed with the rate-limit error. The probe now targets the public GitHub web, which proves connectivity the same way and does not consume API quota.
- "Protected by AdGuard" card showed red at 100% (#959) - the bar applied load-metric thresholds (red above 90%) to a coverage metric, where 100% is the best case. Redesigned: full coverage shows a large green shield with the percentage; partial coverage keeps the protected/total count and the card becomes a toggle that filters the table to the unprotected clients only.
- Duplicate SNMP clients from MAC casing (#960) - the SNMP FDB parser emitted lowercase MACs while every other source (agent probe, wireless, DHCP leases, device attributes) uses uppercase, so the same device appeared twice in the client list; the bridge-MAC uplink exclusion also broke. The parser now emits the canonical uppercase form.
Notes
- No agent changes (still 3.0.5).
Full changelog: CHANGELOG.md
Novedades
Corregido
- El botón de actualizar quedaba bloqueado en instalaciones sin clone git (#955): la comprobación «working tree limpio» se ejecutaba en cualquier instalación en modo rolling (detectado por la presencia de
deploy/update.sh) aunque el directorio no fuera un clone git; fallaba con «no se pudo leer el estado de git» y el botón de aplicar quedaba bloqueado para siempre, pese a queupdate.shsí soporta ese layout. La comprobación se omite cuando no hay.git, igual que en el layout estable. - El chequeo de conectividad agotaba la cuota anónima de la API de GitHub (#956): el sondeo llamaba a
api.github.comen cada cómputo (caché de solo 30 s); con la interfaz de actualización abierta podía superar las 60 peticiones/hora del acceso anónimo y la actualización fallaba después con el error de límite de peticiones. El sondeo usa ahora la web pública de GitHub, que prueba la conectividad igual y no consume cuota de API. - Tarjeta «Protegidos por AdGuard» en rojo al 100 % (#959): la barra aplicaba los umbrales de las métricas de carga (rojo por encima del 90 %) a una métrica de cobertura, donde el 100 % es el mejor caso. Rediseño: con cobertura total muestra un escudo verde grande con el porcentaje; con cobertura parcial muestra el contador de protegidos y la tarjeta es clicable, filtrando la tabla para ver solo los clientes sin proteger.
- Clientes SNMP duplicados por las mayúsculas de la MAC (#960): el parser FDB de SNMP emitía las MACs en minúsculas mientras el resto de fuentes (sondeo del agente, wireless, DHCP, atributos) usan mayúsculas, así que el mismo dispositivo aparecía dos veces en la lista de clientes; también se rompía la exclusión de la MAC del puente en el uplink. El parser emite ahora la forma canónica en mayúsculas.
Notas
- Sin cambios en el agente (sigue en la 3.0.5).
v2.28.41
What's Changed
Added
- WAN / VLANs / alerts series from the borky fork (#946) - multi-WAN support, uplink policy in the agent, alert when the internet exit connection changes, router detail shows its connections, VLANs read from the router, port panel linked by the real port. Attributed.
- Per-device SNMP polling health (#930) - ok/fail counters per SNMP router, unprivileged ICMP ping fallback, snmp-failing alert after 3 consecutive failures only when the host answers ping (hint at a wrong community or port), recovery alert, counters in the router detail and the ports card.
- Last seen for clients (#954) - a new device_seen table tracks the first and last time each MAC was seen online (all sources, not only wireless). The device detail shows both and the clients table gains sortable columns on xl+ widths. The server sends epochs and the app formats the relative time with the active locale.
Fixed
- SNMP switches showed no clients despite a populated FDB (#928) - two chained causes, confirmed with the reporter's snmpwalk: (1) the Q-BRIDGE fallback only ran when the dot1d walk returned zero raw PDUs, so data that did not parse never gave qbridge a chance (#948); (2) TP-Link Omada switches (SG3428X-M2, SG3210X-M2) index the dot1d table by <vlan>.<mac> (7 octets) instead of the plain MAC from RFC 1493, and the parser required exactly 6, dropping every entry (#950). The FDB log line now also shows raw PDU counts per walk.
- "First seen" was never populated outside the demo (#954).
- Auto-generated slugs invisible in the UI (#951) - log lines now lead with the device name (slug in parentheses), and the device detail info card shows an ID row.
Notes
- Agent 3.0.5: the WAN/VLANs series touches the agent (uplink policy); everything else is server and web.
- The clients table column visibility selector (third suggestion from #954) is a follow-up in #958.
- Thanks to @crowedavid for the detailed #928 report, the TP-Link Omada testing and the #954 suggestions, and to borky for the WAN/VLANs series.
Full changelog: CHANGELOG.md
Novedades
Añadido
- Serie WAN / VLANs / alertas del fork borky (#946): soporte multi-WAN, política de uplink en el agente, alerta cuando cambia la conexión de salida a internet, el detalle del router muestra sus conexiones, VLANs leídas del router y panel de bocas enlazado por el puerto real. Atribuido.
- Salud del sondeo SNMP por dispositivo (#930): contadores de aciertos/fallos por router SNMP, ping ICMP de respaldo sin privilegios, alerta snmp-failing tras 3 fallos seguidos solo si el host responde a ping (pista de community o puerto mal configurados), alerta de recuperación y contadores en el detalle del router y en la tarjeta de puertos.
- Last seen en clientes (#954): la nueva tabla device_seen registra la primera y la última vez que el server vio online a cada MAC (todas las fuentes, no solo wireless). El detalle del dispositivo muestra ambas y la tabla de clientes gana columnas ordenables en pantallas xl+. El server manda epochs y la app formatea el relativo con el locale activo.
Corregido
- Switches SNMP sin clientes pese a tener el FDB poblado (#928): dos causas encadenadas, confirmadas con el snmpwalk del reporter. (1) El fallback a Q-BRIDGE solo se probaba cuando el walk dot1d devolvía cero PDUs crudos; datos no utilizables nunca daban opción a qbridge (#948). (2) Los TP-Link Omada (SG3428X-M2, SG3210X-M2) indexan la tabla dot1d por <vlan>.<mac> (7 octetos) en vez de la MAC desnuda de RFC 1493, y el parser exigía exactamente 6: descartaba todas las entradas (#950). La línea de log del FDB muestra además los PDUs crudos de cada walk.
- "First seen" nunca se poblaba fuera del demo (#954).
- Slugs autogenerados invisibles en la UI (#951): las líneas de log usan ahora el nombre del dispositivo (slug entre paréntesis) y la tarjeta de información del detalle muestra la fila ID.
Notas
- Agente 3.0.5: la serie WAN/VLANs toca el agente (política de uplink); el resto son cambios de servidor y web.
- El selector de columnas visibles de la tabla de clientes (tercera sugerencia de #954) queda como follow-up en #958.
- Gracias a @crowedavid por el reporte detallado de #928, las pruebas con sus TP-Link Omada y las sugerencias de #954, y a borky por la serie WAN/VLANs.
Changelog completo: CHANGELOG.md
v2.28.40
What's Changed
Added
- HTTPS with a private certificate authority (#937) - opt-in, defaults unchanged: HTTPS listener with on-demand certificate renewal from Settings, private CA, server key proof at agent pairing (CA pinning, multiple keys), separate HTTP/HTTPS cookies, HSTS on secure responses only, SSE write timeout, downloads through the service worker. Adapted from the borky community fork (17 commits, attributed).
- All / Online / Offline status filter on Clients, persisted (#923) - the online-only switch becomes a three-way selector, saved with the page preferences.
- English as the panel's default language (#938) - new browsers default to English; the switcher is unchanged. From the borky fork.
- Show which AP or switch each client hangs off of (#938).
- Proxmox series from the borky fork (#936) - connection test that reports what the token can see, containers named as Proxmox names them, host matched by the address the cluster knows, VM vs container badges ("VM" on the map), re-classification once a guest gets a name.
- Routers/topology/devices series from the borky fork (#945) - unconfirmed ARP neighbours, client re-attribution when a router is deleted, client counts consistent with the overview, links name the box they come from with measured speeds, unknown-device alert says where it is and offers identification, agent matching by more than its slug, phantom DSA ports gone, uplink marked on the real port.
Fixed
- Reconnecting pill flashing on TLS setups (#925) - only shows after 5 s of sustained SSE outage now.
- Port row overflowing on big switches (#929) - the chassis wraps beyond 16 ports.
- FDB/LLDP clients missing depending on the port (#928, #943) - matching was done by port id instead of interface name; any boca whose id is not its interface showed nothing even with the neighbour learned. Fix from the borky fork, attributed.
- SNMP FDB gave no diagnostics (#928) - the walk now reports its source (dot1d/dot1q), returns a real error when both fail, and the journal logs the entry count on change.
- "Leave anonymous" toasted "Device identified" (#938) - the toast reflects the actual choice.
- "Find this device" got its own label key (#938).
Notes
- Agent 3.0.4: the FDB/LLDP matching and slug series touches the agent; no functional change for routers without those cases.
- Adapted from community forks with attribution: borky (Proxmox, UI/i18n, HTTPS/CA, routers/topology). Open invitations: UniFi (#944), RouterOS (#939).
Thanks to @crowedavid for the reports and to the fork authors.
Full changelog: CHANGELOG.md
Novedades
Nuevo
- HTTPS con autoridad certificadora privada (#937) - opt-in, defaults intactos: listener HTTPS con renovación de certificado desde Ajustes, CA privada, prueba de clave al emparejar el agente (pinning, varias claves), cookies separadas, HSTS solo en seguro, timeout SSE y descargas vía service worker. Del fork borky, atribuido.
- Filtro All / Online / Offline en Clients, persistente (#923).
- Panel en inglés por defecto (#938) - el selector sigue igual.
- A qué AP o switch cuelga cada cliente (#938).
- Serie Proxmox del fork borky (#936) - test de conexión, CTs con su nombre real, host por dirección del cluster, badge "VM" en el mapa.
- Serie routers/topología del fork borky (#945) - ARP no confirmados, enlaces con nombre y velocidad medida, alerta de desconocido localizable e identificable, match de agentes por 3 claves, bocas fantasma DSA fuera, uplink en la boca real.
Arreglado
- Pill "Reconnecting" con TLS (#925) - solo tras 5 s de caída real.
- Chasis de puertos desbordado (#929) - wrap con más de 16 bocas.
- Clientes FDB/LLDP según la boca (#928, #943) - se emparejaba por id de boca, no por interfaz. Fix del fork borky.
- SNMP FDB sin diagnóstico (#928) - fuente (dot1d/dot1q), error real y conteo en el journal.
- Toast "Dejar anónimo" (#938) y clave "Find this device" (#938).
Notas
- Agente 3.0.4 (la serie FDB/slug toca agente).
- Adaptado de forks con atribución: borky (Proxmox, UI/i18n, HTTPS/CA, routers/topología). Invitaciones abiertas: UniFi (#944), RouterOS (#939).
Gracias a @crowedavid por los informes y a los autores de los forks.
Changelog completo: CHANGELOG.md
v2.28.39
What's Changed
Fixed
- Could not select °F in Settings (#898) - the unit was saved with quotes and never matched on read.
- European decimal notation in the WAN totals (#899) - "TOTAL 24H" and "TOTAL TODAY" now follow your locale: the server sends raw bytes, the client formats like every other rate.
- Signal matrix legend did not match the real thresholds (#905) - reworded without relational operators on negative values.
- DAWN filter option on usteer-only networks (#909) - only shown when the network actually has DAWN routers.
- Connect events never appeared (#908) - the logread parser missed the most common hostapd line format (iface listed once); all observed variants parse now.
- Push notifications showed a static 16:00:00 time (#914) - the notifier received events with timestamp 0 (epoch in your timezone); fixed at the source with a regression test.
- "Agent outdated" alert re-fired every 30 s (#910) - it was re-sent on every agent push and refreshed its timestamp; now it fires once per actual version change, like the agent-down alert.
- Initial installer password had ambiguous characters (#895) - no more 0/O or 1/l/I; same entropy (~94 bits).
Added
- Server-wide Weak signal threshold (#904) - the Settings slider now lives on the server and drives the weak-signal alert for every client.
- Matrix "Weak signal only" filter uses that threshold (#906) with an unambiguous label ("weaker than X dBm").
- Configurable roaming events ingest interval (#907) - Settings > Data & thresholds; applies without restart, and the events table header shows your real retention and interval.
- "Unread only" toggle on the Overview alerts card (#916) - keeps the view clean in normal operation.
Thanks to @crowedavid for the relentless, top-quality reports behind most of this release.
Full changelog: CHANGELOG.md
Novedades
Arreglado
- No se podía seleccionar °F en Ajustes (#898) - la unidad se guardaba con comillas y la lectura nunca coincidía.
- Notación decimal europea en los totales WAN (#899) - "TOTAL 24H" y "TOTAL TODAY" siguen tu locale: el servidor manda bytes crudos y el cliente formatea como el resto de tasas.
- La leyenda de la matriz de señal no coincidía con los umbrales reales (#905) - reescrita sin operadores sobre valores negativos.
- Filtro DAWN en redes solo-usteer (#909) - solo aparece cuando la red tiene routers con DAWN.
- Los eventos de conexión no aparecían (#908) - el parser no aceptaba el formato de hostapd más común (iface una sola vez); ahora cubre todas las variantes observadas.
- Los pushes mostraban una hora estática (16:00:00) (#914) - el notifier recibía el evento con timestamp 0 (epoch en tu zona); arreglado en origen con test de regresión.
- La alerta "Agente desactualizado" se repetía cada 30 s (#910) - se reenviaba con cada push y refrescaba su timestamp; ahora dispara una vez por cambio de versión real.
- La contraseña inicial tenía caracteres ambiguos (#895) - sin 0/O ni 1/l/I; misma entropía (~94 bits).
Nuevo
- Umbral "Weak signal" server-wide (#904) - el slider de Ajustes vive ahora en el servidor y alimenta la alerta de señal débil para todos los clientes.
- El filtro "Solo señal débil" de la matriz usa ese umbral (#906) con rótulo inequívoco ("peor que X dBm").
- Intervalo de ingesta de eventos de roaming configurable (#907) - Ajustes > Datos y umbrales; sin reinicio, y la cabecera de la tabla muestra tus valores reales.
- Toggle "Solo no leídas" en la tarjeta de Alertas del Overview (#916) - vista limpia en operación normal.
Gracias a @crowedavid por los informes impecables detrás de casi todo este release.
Changelog completo: CHANGELOG.md
v2.28.38
What's Changed
Fixed
- "Mark all as read" did not stick (#891) — a snapshot taken before the request reached the server could bounce alerts back to unread until the next cycle. The read state is now re-applied once the operation confirms, and SSE alert events are deduplicated by ID.
- Push notifications were always sent in Spanish (#888) — alerts are composed in Spanish and the web UI translates them per user, but ntfy, Telegram and webhook published the raw text. They are now translated server-side at publish time using the same app catalogs.
- "Agent down - reinstall" banner right after a server restart (#887) — for the first 90 s after startup, router cards now show a neutral "Waiting for agents to reconnect…" banner; the drastic banner is reserved for persistent failures.
- Agent env parser kept inline comments (#883) —
NETPULSE_SCAN_INTERVAL=0 # notewas read as the literal value and fell back to the 30 m default instead of disabling the scan. Agent 3.0.3. - Updates failed on installs without a local git clone (#897) — in the stable layout the update died at the fetch step; the main SHA is now resolved via
git ls-remote.
Added
- Server-wide notification language (#889) — new
alerts.langsetting (Settings > Notifications, default English) for ntfy, Telegram and webhook alerts, independent of each client's UI language. - Agent recovery logging (#890) — the whole automatic recovery flow (TTL expiry, SSH restart, push wait, token rotation, reinstall escalation, cooldowns) is now logged with
supervisor:/rearmer:prefixes, and the full strategy is documented indocs/AGENTE-OPENWRT.md.
Full changelog: CHANGELOG.md
Novedades
Arreglado
- "Marcar todo como leído" no persistía (#891) — un snapshot tomado antes de que la petición llegara al servidor devolvía las alertas a no leído. Ahora el estado se re-aplica al confirmarse y los eventos SSE se deduplican por ID.
- Los pushes salían siempre en español (#888) — las alertas se componen en español y la web las traduce por usuario, pero ntfy, Telegram y webhook publicaban el texto en crudo. Ahora se traducen en el servidor al publicar.
- Banner de "agente caído" tras un reinicio (#887) — durante los primeros 90 s tras el arranque se muestra un aviso neutro de espera; el banner drástico se reserva para fallos persistentes.
- El parser del env del agente no recortaba comentarios en línea (#883) — agente 3.0.3.
- La actualización fallaba sin clone git (#897) — el SHA de main se resuelve ahora vía
git ls-remote.
Nuevo
- Idioma por defecto del servidor para las notificaciones (#889) — ajuste
alerts.lang(Ajustes > Notificaciones, inglés por defecto), independiente del idioma de cada cliente. - Log de la recuperación de agentes (#890) — todo el proceso queda registrado con prefijos
supervisor:/rearmer:y documentado endocs/AGENTE-OPENWRT.md.
Changelog completo: CHANGELOG.md
v2.28.37
What's new
Fixes
- Agent reinstall no longer wipes your config (#851): the console reinstall and the hot token rotation rewrote
/etc/netpulse-agent.envfrom scratch and lost hand-added settings likeNETPULSE_SCAN_INTERVAL=0. Both now preserve any option they don't manage, and the generated env documents the available options as commented defaults. - Watchdog cron removed from new installs (#851): a blind restart every 2 minutes doesn't fix a crash loop, it just spams the log; the server's dead-man detection plus alerts is what surfaces the problem. procd respawn still covers one-shot crashes, the init restores the binary after sysupgrade, and upgrading cleans the old cron entry and watchdog file.
- Stale watchdog cron entry self-heals after sysupgrade (#879):
/etcsurvives but the binary doesn't, so cron logged a missing-command error every 2 minutes. The agent init now removes the orphaned entry at start. - Alert actions existed only on hover (#862): on phones and tablets they were unreachable. Touch devices now show them always, with their label on its own line so the title keeps its width.
- CI injected the agent tag as the server version (#869): dev builds reported
agent/vX.Y.Z; only release tags are considered now.
Added
- Per-switch console polling toggle (#863): the RTLPlayground firmware console allows a single session and every poll kicked whoever was using the web UI. You can now disable polling per switch (Settings > Routers); with beacon v1.3+ firmware, uptime, MAC and firmware still arrive over UDP.
- Per-channel notification verbosity (#874): each channel (ntfy today, same switch for Telegram, webhook and Web Push) decides whether it receives urgent alerts only (default, unchanged) or everything that passes the feed filter.
- Notification channels documented (#871): the ntfy and Telegram cards and the README now state that channels carry urgent alerts only, separate from the feed level.
Novedades
Correcciones
- Reinstalar el agente ya no pisa tu configuración (#851): el reinstall de la consola y la rotación de token reescribían el env desde cero y perdían ajustos añadidos a mano. Ahora conservan lo que no gestionan y el env documenta las opciones disponibles.
- Watchdog cron fuera de la instalación (#851): un reinicio a ciegas no arregla un bucle de crash; la detección real es la del servidor. El respawn de procd cubre crashes puntuales y el init restaura el binario tras sysupgrade.
- La entrada de cron del watchdog huérfana se auto-repara tras sysupgrade (#879): el init del agente retira la entrada al arrancar si el watchdog ya no existe.
- Las acciones de alertas solo existían con hover (#862): en táctil se muestran siempre, con etiqueta en su propia línea.
- El CI inyectaba el tag del agente como versión del servidor (#869): ahora solo se consideran tags de release.
Añadido
- Toggle "Sondeo por consola" por switch (#863): la consola del firmware RTLPlayground admite una sola sesión y cada sondeo echaba a quien estuviera dentro. Desactívalo por switch; con el beacon v1.3+ los datos llegan por UDP.
- Verbosidad por canal de notificación (#874): cada canal decide entre solo urgentes (por defecto) o todo lo que pasa el filtro del feed.
- Canales de notificación documentados (#871): tarjetas de ntfy/Telegram y README explican que los canales solo llevan alertas urgentes.
v2.28.36
What's new
Fixes
- Agent reinstall over HTTPS (#851): the console reinstall (manual and the supervisor's auto-reinstall) generated
/etc/netpulse-agent.envwithoutNETPULSE_SERVER_FP, so the agent kept crash-looping on the SPKI pinning check even with a valid TLS certificate. The server now derives the SPKI pin itself (TLS handshake against its own public URL, cached) and the generated env carries it; the hot token rotation preserves it instead of dropping it. - Alert actions easier to see (#862): feed row actions are larger (bigger icon, more generous hit area) and show their text label on hover; the Home summary reduces each alert to its title with the same actions, and "Update agent" now goes to the fleet's agent section.
- Custom device icon missing in the Connected Clients table (#864): the table used the deduced type and ignored
iconOverride; it now follows the same rule as the rest of the panel.
Added
- Firmware, uptime and MAC via beacon (#865): the RTLPlayground firmware status datagram now carries firmware version, uptime and switch CPU MAC, and NetPulse ingests them as the primary source (with console fallback), so the panel shows them even with console polling disabled.
Novedades
Correcciones
- Reinstalar agente desde la consola con servidor HTTPS (#851): el script de reinstall (manual y el auto-reinstall del supervisor) generaba el
/etc/netpulse-agent.envsinNETPULSE_SERVER_FPy el agente seguía cayendo en el bucle fatal de pinning aunque el TLS fuera válido. El servidor deriva ahora el pin SPKI por sí mismo (handshake TLS contra su propia URL pública, con caché) y el env generado lo lleva; la rotación de token en caliente también lo conserva. - Acciones de alertas más visibles (#862): las acciones por fila del feed crecen (icono mayor y área de clic más generosa) y muestran su etiqueta al pasar el cursor; el resumen de Inicio reduce cada alerta al título con las mismas acciones, y "Actualizar agente" navega a la sección de agentes de la flota.
- El icono personalizado de un cliente no se veía en la tabla de Clientes (#864): la tabla usaba el tipo deducido e ignoraba
iconOverride; ahora sigue el mismo criterio que el resto del panel.
Añadido
- Firmware, uptime y MAC por beacon (#865): el datagrama de estado del firmware RTLPlayground lleva ahora versión de firmware, uptime y MAC de la CPU, y NetPulse los ingiere como fuente principal (con fallback a consola), así que el panel los muestra aunque el sondeo por consola esté deshabilitado.
v2.28.35
What's new / Novedades
EN
Reliability of agent monitoring and fleet alerts:
- Agent-only routers no longer flap offline on every client roam: the registry retains the agent identity across wireless-only event pushes (#852).
- Agent-down on agent-only routers is now an urgent router alert, so it passes urgent-only notification filters (#850).
- One offline alert per outage: no re-emits while flapping, clean removal from the feed on stable recovery, survives restarts (#846).
- Fresh agent installs against an HTTPS server work out of the box: the installer derives the SPKI fingerprint from the certificate itself (#851).
UI and ports:
- The contracted WAN speed now shows on the topology link (#849).
- 2.5 GbE links display as "2.5 Gbps" and internal DSA conduits no longer appear as fake ports (#847). Agent-side: requires the updated agent (3.0.1) on the routers.
- Custom device icons are honored on the topology map chips (#845).
- Fleet table matches agents by slug or routerId (#848).
Landing:
- One crawlable route per language (10 languages x 3 pages) with proper lang, canonical, hreflang and og:locale; the sitemap lists all 30 URLs (#842).
- Native review pass of the ar, hi, zh and ja translations (#841).
- README documents the anonymous instance counter and its opt-out (#822).
ES
Fiabilidad del monitoreo por agente y de las alertas de flota:
- Los routers solo-agente ya no flapean offline con cada roam: el registry conserva la identidad del agente entre pushes de eventos wireless (#852).
- La caida del agente en routers solo-agente ahora es alerta urgente de router y pasa los filtros de notificacion (#850).
- Una alerta offline por caida: sin re-emisiones durante el flapping, se retira del feed al recuperarse de forma estable y sobrevive a reinicios (#846).
- Instalaciones de agente contra un servidor HTTPS funcionan solas: el instalador deriva el fingerprint SPKI del propio certificado (#851).
Interfaz y puertos:
- La velocidad WAN contratada se ve ya en el enlace de la topologia (#849).
- Los enlaces de 2.5 GbE se muestran como "2.5 Gbps" y los conduits DSA internos dejan de aparecer como bocas falsas (#847). En el router hace falta el agente actualizado (3.0.1).
- El icono personalizado de un cliente se respeta en los chips de la topologia (#845).
- La tabla de flota empareja agentes por slug o routerId (#848).
Landing:
- Una ruta indexable por idioma (10 idiomas x 3 paginas) con lang, canonical, hreflang y og:locale correctos; el sitemap lista las 30 URLs (#842).
- Repaso nativo de las traducciones ar, hi, zh y ja (#841).
- El README documenta el contador anonimo de instancias y su opt-out (#822).
Full details in CHANGELOG.md.
v2.28.34
Changelog
- caf3de475a3e67fd2584c187c2f2e78e1f9402dd chore: prepare v2.28.34
- babc6efc4b9c899676cccc7ebc65d04489557f97 docs(web): complete the eight non-ES/EN translations and read the hl parameter (#840)
- 07f2110bb8ff0174a4cb0caded5fb937093fbb78 feat(mqtt): settings card and propagation to NetGrip routers (#843)
- 2d48b0fa253a86457367b1d08e7e19681a51a0e2 feat: alert actions, NetGrip card in orchestration, about links, MQTT settings (#844)
v2.28.33
Changelog
- 4e53b4160a5fc0080122c00941f260e93f2e7b12 chore(web): bump the asset cache-busting version and add the new page to the sitemap
- af8fc8243c147e6599b411048040a22d7b86390c chore: prepare v2.28.33
- 789a6bfbd89544bce39d09a6ef737281cc1af532 feat(alerts): handle unknown-device alerts directly and edit clients inline (#829)
- 5b36374486376fa13c78441f2b0d63d8bc7e541d feat(mqtt): publish fleet state to an MQTT broker with Home Assistant discovery (#826)
- 0bf1f2ba0fb184e8f8aed10f43552e335311c8a2 feat(telemetry): anonymous daily instance ping (#822) (#824)
v2.28.32
Changelog
- 86984994313a1fe41b2d127543f427ee22eff375 chore(announcements): publish the GitHub stars notice
- 99b979124f21f9087d1c24911bfb877f6c3e7f26 chore: prepare v2.28.32
- 38e98e74a7316a6b0d2d407e5e20021af5174f7a fix(backup): compress scheduled database backups (#818) (#820)
- b38995af75c39479966ccc6e5e66dd5a1387db98 fix(db): drop redundant port_series_raw indexes duplicating the PK (#817) (#819)
- f73f477b88a053665f447753256544c43adfd6d0 fix(pwa): let the OS manage screen orientation instead of forcing any (#808) (#823)
- d53bb4355e9c181fe9aa19e4d435ec597df27e09 fix(retention): size raw rollup window to the raw retention (#812) (#821)
- 5d61eb51ef504edc1d4a8d1aab5aab6f649fc17b fix(updater): prune old .update-backup-* snapshots (#809) (#815)
- ea423beeb03be10d435dd37313e30a87bf6122de test(httpapi): make backup-loop test deterministic (#807) (#816)
v2.28.31
Changelog
- 287b4ca85995cd857ab720978b27a714a75381b0 chore(announcements): serve the feed from netpulse.cloudless.club (#810) (#811)
- f75050c48d9c3b398c5a18b5b6e65bc3fc423857 chore: prepare v2.28.31
- 9586a273cebea24c838fb0eda394898cfc45564c fix(settings): make the Services toggle control AdGuard monitoring (#814)
v2.28.30
What's new
Alert log history survives restarts (#798) — the alert log was memory-only, so a self-update wiped it. Alerts are now persisted to SQLite and reloaded on boot.
Apply a device's name as a DHCP reservation (#800) — the device edit sheet can now push the display name to the router: if the MAC already has a reservation only its hostname is updated, otherwise the reservation is created with the current IP. DNS-valid hostnames only, IP conflicts rejected, and like every router write it goes through the dry-run command plan before running.
Editable name and type per client (#797) — the edit sheet gains a name field and a manual type selector (persisted as overrides), fixing rename-only-icon and the type filter falling back to "Unknown" too often.
Stable sorting and a traffic column (#799) — the default order no longer jumps between refreshes, the desktop list has a sortable Traffic column and mobile gets a sort control.
The update check button runs a real check (#795) — it now triggers a fresh check instead of only reading the cached status.
Auto-update alerts respect the UI language (#796) — the alert now resolves from the EN/ES locales using stable result codes.
Novedades
El historial de alertas sobrevive a reinicios y actualizaciones (#798) — el registro era solo memoria y un self-update lo borraba; ahora se guarda en SQLite y se recarga al arrancar.
El nombre del dispositivo se puede aplicar como reserva DHCP (#800) — el editor puede empujar el nombre visible al router: si la MAC ya tiene reserva se actualiza solo el hostname; si no, se crea con la IP actual. Solo hostnames DNS válidos, conflictos de IP rechazados y siempre con plan de comandos antes de ejecutar.
Nombre y tipo editables por cliente (#797) — el editor gana campo de nombre y selector de tipo manual (guardados como overrides).
Orden estable y columna de tráfico (#799) — el orden por defecto ya no baila entre refrescos, la lista de escritorio tiene columna de Tráfico y el móvil un selector de orden.
El botón de comprobar actualizaciones hace un check real (#795) — ahora dispara el check de verdad en vez de leer solo el estado cacheado.
La alerta de auto-actualización respeta el idioma de la interfaz (#796) — se resuelve desde los locales EN/ES con códigos de resultado estables.
v2.28.28
What's Changed
Agent versioning decoupled from the server release (#790). Until now every server release rebuilt the agent with the release tag number, so the whole fleet was flagged as outdated on every release even when the agent code had not changed at all. The agent now has its own version, stored in agent/VERSION (starting at 3.0.0), which CI injects into both the agent binary and the server. The fleet updates once to 3.0.0 and from then on only gets an update prompt when the agent code actually changes; the release workflow fails if the agent code changed without a version bump. Updating from older versions jumps straight to this release - database migrations are state-based and idempotent, no step-by-step path needed.
The rearm button is no longer offered for external agents (#792). The router detail header showed "Rearm" for any stale agent, including external pushers without SSH (e.g. a beacon-managed switch), which the backend already rejected. The button now hides itself for external kinds everywhere.
Qué hay de nuevo
Versionado del agente desacoplado del server (#790). Hasta ahora, cada release del server rebuildaba el agente con el número del tag y toda la flota se marcaba como desactualizada en cada release aunque el código del agente no hubiera cambiado. El agente pasa a tener versión propia en agent/VERSION (arranca en 3.0.0), que el CI inyecta tanto en el binario del agente como en el server. La flota se actualiza una vez a 3.0.0 y, a partir de ahí, solo recibe aviso de actualización cuando el agente cambia de verdad; además, la release falla si el código del agente cambió sin subir la versión. Actualizar desde versiones anteriores salta directo a esta release: las migraciones de la base de datos son por estado e idempotentes, sin pasos intermedios.
El botón «Rearmar» ya no se ofrece en agentes externos (#792). La cabecera del detalle de un router mostraba el rearme para cualquier agente caído, incluidos los externos sin SSH (p. ej. un switch gestionado por beacon), que el backend ya rechazaba. Ahora el botón se oculta para tipos externos en todos los sitios.
Full changelog: https://github.com/gnacho/netpulse/blob/main/CHANGELOG.md
v2.28.26
NetPulse v2.28.26
New
- Full firmware upgrade flow with owut/ASU (#761): the upgrade screen is rebuilt around attended sysupgrade — target version dropdown with the stable builds ASU can produce, in-place owut install, one-shot/recurring schedules, progress and success views, GL.iNet vendor guard, and stack packages preserved across flashes.
- ntfy notifications (#766) with security hardening (#773): topic treated as a secret (masked in logs/errors), tighter retries, rune-safe truncation, idempotent close.
- Proxmox VE multi-instance (#764): several clusters or standalone nodes, legacy config auto-migrates, secrets never leave the server.
- Scheduled self-update (#759): daily/weekly/monthly opt-in with verified apply and rollback.
- TLS certificate hot-reload (#769): renewed certs (acme.sh/certbot) are served without restarting.
- Unknown-device onboarding (#772): the alert now notifies through your channels and deep-links to an intake card — name, icon, optional IP reservation with a dry-run command plan, or leave it anonymous to silence it.
- Per-client presence timeline (#771): 7-day presence bands colored by AP in the device detail, recent event feed, fleet roaming-health card, configurable retention with hourly pruning.
- Topology view persists zoom/pan (#777) and the Clients list persists view mode and sort (#778).
Fixed
- Health subscore chip and unknown OUI manufacturer now follow the UI language (#770).
- AmneziaWG interfaces are detected via the awg tool (#776).
- RTL console uptime parses the new tick/sec counter format (#785).
- "Check for updates" no longer claims you are up to date when the check itself failed (already in v2.28.24 line, plus all fixes from v2.28.24).
NetPulse v2.28.26 (español)
Novedades
- Flujo completo de actualización de firmware con owut/ASU (#761): desplegable de versiones que ASU puede construir, instalación de owut desde la UI, programación puntual/recurrente, vistas de progreso y éxito, bloqueo de firmware de fabricante GL.iNet y preservación de los paquetes del stack en el flasheo.
- Notificaciones vía ntfy (#766) con endurecimiento de seguridad (#773): el topic se trata como secreto, reintentos más ajustados, truncado seguro y cierre idempotente.
- Proxmox VE multi-instancia (#764): varios clusters o nodos sueltos, migración automática de la config anterior.
- Auto-actualización programada del servidor (#759): diaria/semanal/mensual, opt-in, con verificación y rollback.
- Recarga del certificado TLS sin reiniciar (#769): las renovaciones de acme.sh/certbot se sirven al instante.
- Alta guiada de dispositivos desconocidos (#772): la alerta notifica por tus canales y enlaza a una tarjeta de alta (nombre, icono, reserva IP con plan de comandos, o dejarlo anónimo).
- Timeline de presencia por cliente (#771): bandas de 7 días coloreadas por AP, feed de eventos, tarjeta de salud de roaming y retención configurable con poda horaria.
- La topología recuerda el zoom (#777) y la lista de clientes recuerda vista y orden (#778).
Correcciones
- El chip "Servicios" y el fabricante desconocido se muestran ya en el idioma de la UI (#770).
- Las interfaces AmneziaWG se detectan con la herramienta awg (#776).
- El uptime de la consola RTL parsea el nuevo formato de contadores (#785).
v2.28.24
EN
Honest update checks (#743): when the check cannot reach GitHub (anonymous API rate limit, network hiccup), the button used to show the last known state as "You're on the latest version". It now says "Could not check right now (reason); try again in a few minutes" instead of lying.
Documentation honesty (#751, #752): the project grew from watching into also writing (DHCP reservations, device blocks, orchestration, firmware upgrades), so every blanket "read-only" claim in the READMEs, website, manuals and app strings was replaced with the accurate split: polling is read-only; anything that writes is explicit, admin-triggered, and covered by a snapshot with automatic rollback. English UI now consistently says "Firmware upgrades".
Dry-run command preview (#754): reserving an IP or blocking a device no longer applies directly. The server first returns the exact commands that would run on the router (and the rollback commands), the device sheet shows them, and nothing is executed until you press "Run and apply".
External announcements channel: the server fetches announcements.json from the repository and the UI renders community-wide notices in the same style as the update ribbon, with a large 22/18 px type, optional link and a persistent per-user dismiss. First notice: the NetGrip beta tester program is open.
As usual, the built-in updater offers this release and applies it with an atomic swap and automatic rollback if the service does not come back.
ES
Comprobación de actualizaciones honesta (#743): si el check no llega a GitHub (límite de la API anónima, red), el botón devolvía el último estado conocido como "estás en la última versión". Ahora dice "No se pudo comprobar ahora (motivo); reintenta en unos minutos" en lugar de mentir.
Honestidad documental (#751, #752): el proyecto pasó de mirar a también escribir (reservas DHCP, bloqueos, orquestación, firmware), así que todo claim genérico de "solo lectura" en READMEs, web, manuales y textos de la app se sustituye por el matiz correcto: el sondeo es de solo lectura; lo que escribe es explícito, lo lanza un admin y va con snapshot y rollback. El UI en inglés unifica "Firmware upgrades".
Previsualización de comandos (dry-run) (#754): reservar una IP o bloquear un dispositivo ya no aplica directo. El servidor devuelve primero los comandos exactos que se ejecutarían en el router (y los de rollback), la ficha los muestra, y no se ejecuta nada hasta pulsar "Ejecutar y aplicar".
Canal de avisos externos: el servidor consulta announcements.json del repositorio y la UI pinta avisos para toda la comunidad con el estilo del ribbon de actualizar, en cuerpo grande 22/18 px, enlace opcional y descarte persistente por usuario. Primer aviso: el programa de beta testers de NetGrip está abierto.
Como siempre, el actualizador integrado ofrece esta release y la aplica con swap atómico y rollback automático si el servicio no vuelve.
v2.28.22
EN
- Alert localization, complete (#712, #689): the alert feed and the web push notifications now translate from the alert type and its vars in the device language, instead of showing server-side Spanish literals.
- Per-router high-temperature threshold (#716): each router can set its own threshold (default 65 C), used by the alert and the hot state. The global slider, which never governed the server alert, is gone.
- Celsius or Fahrenheit (#717): a display preference applied across the UI and the in-app temperature alerts.
- Configurable router poll interval (#715):
NETPULSE_POLL_INTERVAL(default 30 s, previously 5 s hardcoded), and the UI refresh setting now really drives the client fallback instead of pretending to change the server poll. - All WireGuard interfaces (#713, #714): the poll discovers every
wginterface on the router, and a single failing interface no longer takes the whole panel down. - Device grid density (#711): more columns on wide screens, so more devices fit per view.
- Topology (#692, #694): labels no longer overlap the device icons, and quiet clients of a bridge AP stay attributed to the AP.
- Relative times (#698, #691): the alerts summary shows the real relative time and the agent last-seen cell ticks every second.
- Attach overrides to a router or port (#690): a manual override can target a router id or
routerId:port, not only a device MAC. - Optional HTTPS listener (#696):
NETPULSE_TLS_ENABLEDserves HTTPS with your certificates, or a self-signed fallback, without touching the HTTP the agents use. - owut check (#695): detect
owuton a router and runowut checkfor review before any upgrade. - Security (#693): the DHCP reservation hostname is validated and escaped before it reaches a router command.
- Clearer agent actions (#718, #719): rearm shows the real reason, and token rotation no longer tells you to reinstall an agent when there is none.
- Landing (#734): the about copy is reworked and the Ko-fi ask is replaced by asking for a GitHub star.
ES
- Localización de alertas, completa (#712, #689): el feed de alertas y las notificaciones push se traducen ahora desde el tipo de alerta y sus variables en el idioma del dispositivo, en vez de mostrar literales en español del servidor.
- Umbral de temperatura alta por router (#716): cada router puede fijar su propio umbral (65 C por defecto), usado por la alerta y el estado caliente. El slider global, que nunca gobernaba la alerta del servidor, desaparece.
- Celsius o Fahrenheit (#717): preferencia de presentación aplicada a toda la interfaz y a las alertas de temperatura en la app.
- Intervalo de sondeo configurable (#715):
NETPULSE_POLL_INTERVAL(30 s por defecto, antes 5 s fijos), y el ajuste de refresco de la interfaz ahora gobierna de verdad el respaldo del cliente en vez de aparentar cambiar el sondeo del servidor. - Todas las interfaces WireGuard (#713, #714): el sondeo descubre todas las interfaces
wgdel router, y una interfaz que falla ya no tumba el panel entero. - Densidad de la vista de dispositivos (#711): más columnas en pantallas anchas, para ver más dispositivos de una vez.
- Topología (#692, #694): los nombres ya no se montan sobre los iconos, y los clientes callados de un AP puente se quedan atribuidos al AP.
- Tiempos relativos (#698, #691): el resumen de alertas muestra el tiempo relativo real y la celda "Visto" de los agentes avanza segundo a segundo.
- Overrides attach a router o puerto (#690): un override manual puede apuntar al id de un router o a
routerId:puerto, no solo a la MAC de un dispositivo. - Listener HTTPS opcional (#696):
NETPULSE_TLS_ENABLEDsirve HTTPS con tus certificados, o con un autofirmado de reserva, sin tocar el HTTP que usan los agentes. - owut check (#695): detecta
owuten un router y ejecutaowut checkpara revisión antes de cualquier actualización. - Seguridad (#693): el hostname de la reserva DHCP se valida y se escapa antes de llegar a un comando del router.
- Acciones de agente más claras (#718, #719): el rearm muestra el motivo real y la rotación de token ya no te dice que reinstales un agente cuando no hay ninguno.
- Landing (#734): el copy del about se reescribe y la petición de Ko-fi se sustituye por pedir una estrella en GitHub.
v2.28.21
EN
/api/healthnow counts agent SSE connections (#681):sseConnectionsonly reflected open browser tabs; agents connect through their own hub and were invisible. A newagentSseConnectionsfield reports them, so a fleet fully connected no longer reads as 0-1.
ES
/api/healthcuenta ahora los SSE de los agentes (#681):sseConnectionssolo reflejaba las pestañas abiertas; los agentes conectan por su propio hub y quedaban invisibles. El nuevo campoagentSseConnectionslos reporta: una flota totalmente conectada ya no lee 0-1.
v2.28.20
EN
- Quiet wired clients stay on their AP (#678): a device behind a bridged AP that stops talking no longer drifts to the main router; the server remembers where the cable actually is, counts only local sightings, and never invents presence.
- VPN tunnel lines follow the nodes (#677): rearranging the map no longer leaves WireGuard curves pointing at empty space.
- The agent self-recovers after a long outage (#680): stale buffered payloads are dropped instead of deadlocking the queue with anti-replay 401s.
ES
- Los clientes cableados callados se quedan en su AP (#678): un dispositivo tras un AP puente que deja de hablar ya no se va al router principal; el servidor recuerda dónde está el cable, solo cuenta observaciones locales y no inventa presencia.
- Las líneas de los túneles VPN siguen a los nodos (#677): reordenar el mapa ya no deja curvas WireGuard apuntando al vacío.
- El agente se recupera solo tras un corte largo (#680): los payloads caducados del buffer se descartan en vez de bloquear la cola con 401 anti-replay.
v2.28.19
EN
- Alerts render in your language (#671): titles, descriptions and hints carry their type and values, and the app translates them in the active language — no more mixed Spanish in an English feed.
- Hostnames survive the browser translator (#666): the page language now follows the UI and network data is marked do-not-translate, so translators stop "correcting" names like crowed-pixeltab.
- OpenWrt packages for x86/64 (#672): agent and on-router server published for x86/64 (.ipk 24.10, .apk 25.12), architecture derived from the SDK.
- Dev dependency bump: js-yaml 4.3.2.
ES
- Las alertas en tu idioma (#671): título, descripción y sugerencia llevan su tipo y sus valores, y la app los traduce al idioma activo — adiós al feed mezclado en español con la app en inglés.
- Los hostnames sobreviven al traductor del navegador (#666): el idioma de la página sigue a la UI y los datos de red se marcan como no traducibles, así el traductor deja de "corregir" nombres como crowed-pixeltab.
- Paquetes OpenWrt para x86/64 (#672): agente y servidor on-router publicados para x86/64 (.ipk 24.10, .apk 25.12), arquitectura derivada del SDK.
- Bump de dependencia de desarrollo: js-yaml 4.3.2.
v2.28.18
EN
Topology you can trust (and finally adjust)
- Wired hosts of a bridged dumb AP show under that AP instead of the main router. If a node bridges into the main LAN, its own Ethernet ports are the most specific evidence of where a cable goes, so those clients stay on the AP. The uplink keeps being excluded as transit.
- The map stops jumping around: device order is deterministic, and quiet devices (TVs, receivers, media players) keep their switch placement thanks to a 30-minute port memory that only backs up the live FDB.
- Edit mode (admin): drag routers, switches and devices to adjust distances; save and the layout stays frozen. Reset returns to the automatic arrangement.
- Proxmox hosts render as round nodes with their CTs/VMs nested underneath (+N badge), and dragging a host moves its whole VM family.
- Device card: discreet inline Tag button (no MAC typing), 1.2s travel grace, formatted traffic. The packet-flow toggle is gone; flow follows the reduce-motion preference.
ES
Topología fiable (y por fin ajustable)
- Los hosts por cable de un dumb AP puentado cuelgan de ese AP en vez del router principal: sus bocas Ethernet son la evidencia más específica de por dónde va el cable. El uplink sigue excluido como tránsito.
- El mapa deja de saltar: orden de dispositivos determinista y los dispositivos callados (TVs, receptores, reproductores) conservan su switch gracias a una memoria de puerto de 30 min que solo respalda el FDB vivo.
- Modo edición (admin): arrastra routers, switches y dispositivos para ajustar distancias; al guardar, el layout queda congelado. "Restablecer" vuelve al automático.
- Los hosts de Proxmox se dibujan como nodos redondos con sus CTs/VMs anidados debajo (badge +N), y arrastrar el host mueve a toda su familia.
- Tarjeta de dispositivo: botón discreto Etiquetar (sin teclear MAC), 1,2s de gracia para llegar a ella y tráfico formateado. Fuera el toggle "Flujo": la animación sigue el ajuste de reducir animaciones.
v2.28.17
Changelog
- 22efff302fb48e4d49d6e3556eb6e002228c0a91 chore: prepare v2.28.17 release (#668)
v2.28.16
Changelog
- 469af14f645e0f0d61e1087d3ebe6d7fd6ee0661 chore: prepare v2.28.16 release (#665)
v2.28.15
Changelog
- 902516d3497b4997ce5882c9ef4514728d4cfa1a chore: prepare v2.28.15 release (#652)
v2.28.14
Changelog
- c79941f143561d50bb38db59f1b9f50b2b839d2b chore: prepare v2.28.14 release (#638)
v2.28.13
Changelog
- 3b24255651ad27ba0aa9242be404af3614853320 chore: prepare v2.28.13 release (#633)
v2.28.12
Changelog
- c5a5198e3651c2279aad622d9c4d6aecfd3e3a48 chore: prepare v2.28.12 release (#620)
v2.28.11
v2.28.11
English
Fixed
- Agent degraded AP WiFi (#591): the agent ran an active
iw dev scan(which takes the radio off channel) on every full probe, ~every 30-37 s per device (measured in the scan table). On access points this repeatedly disconnects IoT clients and causes signal loss. Passive scans now run at most every 10 minutes, after startup, or immediately when the server requests a refresh over SSE. Embedded agent version bumped to 2.26.12 so the fleet can update. Also fixes a map-order flake inTestParseUsteerthat broke CI. - SSH re-onboard silently accepted changed keys (#603): on a known host with a different key, the pool refreshed
known_hostsin the background (accept-new with refresh), defeating MITM protection. The pool now rejects the connection and marks the routerunreachable/hostKeyChanged; an admin confirms the re-onboard explicitly (new endpointPOST /api/routers/{id}/accept-host-keyand a button in the router detail), which drops the known_hosts entry and re-runs TOFU on the next poll. The recorded key type is whatever the router's dropbear presents (ssh-rsa in our OpenWrt fleet); NetPulse does not force RSA.
Español
Corregido
- El agente degradaba el WiFi de los APs (#591): lanzaba un
iw dev scan(scan activo que saca la radio del canal) en cada sondeo completo, cada ~30-37 s por dispositivo (medido en la tabla de scans). En puntos de acceso provoca desconexiones de clientes IoT y pérdida de señal. Ahora se ejecuta como mucho cada 10 minutos, tras el arranque, o de inmediato si el servidor pide un refresh por SSE. La versión de agente embebida sube a 2.26.12 para que la flota se actualice. También corrige un test flaky de orden de mapa enTestParseUsteerque rompía el CI. - El re-onboard SSH auto-aceptaba claves cambiadas (#603): ante una clave distinta en un host conocido, el pool renovaba
known_hostsen silencio (accept-new con refresh), anulando la protección MITM. Ahora la conexión se rechaza y el router se marca comounreachable/hostKeyChanged; un admin confirma el re-onboard explícitamente (nuevo endpointPOST /api/routers/{id}/accept-host-keyy botón en el detalle del router), que borra la entrada de known_hosts y vuelve a hacer TOFU en el siguiente sondeo. El tipo de clave registrado es el que presente el dropbear del router (ssh-rsa en nuestra flota OpenWrt); NetPulse no fuerza RSA.
v2.28.10
v2.28.10
English
Fixed
- The router detail RAM caption always showed "512 MB total" (#590): it was computed from the demo extras resolver (whose fallback for any unknown router is the demo entry with 512 MB), even in live mode. It now uses the real
ramMbfrom the backend (/api/routers/{id}).
Added
- Live performance history on the router detail: the backend already served the real 1h/24h/7d cpu/ram/temp series but the front ignored them and painted the synthetic demo series (always "peaking at 21:00", random walk around the current value). In live mode the three charts now use the real per-router history for the selected range; demo keeps its canonical series. A device without history yet shows a single point with its current value instead of breaking the panel.
Español
Corregido
- El caption de RAM del detalle de router siempre mostraba "512 MB total" (#590): se calculaba con el resolver de extras de demo (cuyo fallback para cualquier router desconocido es la entrada demo con 512 MB), incluso en live. Ahora usa el
ramMbreal del backend (/api/routers/{id}).
Añadido
- Historia real de rendimiento en el detalle de router: el backend ya servía las series reales 1h/24h/7d de cpu/ram/temp pero el front las ignoraba y pintaba la serie sintética de demo (siempre "picaba a las 21:00", recorrido aleatorio alrededor del valor actual). En live las tres curvas usan ahora la historia real por router y rango; la demo conserva su serie canónica. Un dispositivo sin historia todavía muestra un punto con su valor actual en lugar de romper el panel.
v2.28.9
v2.28.9
English
Changed
- The gateway card on the "Devices" view is now identified by role (
roleBadge Principal) instead of the demo-onlyflint2id, so in live mode it receives the primary router styling (accent tile, AdGuard/WireGuard chips). Both the fleet card and the router detail header now show the realrouter.modelinstead of the hardcoded "GL.iNet Flint 2 · GL-MT6000" caption, and the unusedmsToGatewayi18n key was removed (#597).
Español
Cambiado
- La tarjeta del gateway en "Dispositivos" ahora se identifica por rol (
roleBadge Principal) en vez del id de demoflint2, así que en modo live recibe el tratamiento de router principal (tile en acento, chips AdGuard/WireGuard). Tanto la tarjeta como la cabecera del detalle muestran elmodelreal del router en lugar del literal "GL.iNet Flint 2 · GL-MT6000", y se elimina la clave de traducción sin usomsToGateway(#597).
v2.28.8
v2.28.8
English
Fixed
- Fleet cards on the "Devices" view no longer label every router as "Main gateway". In live mode the footer fell back to that label because gateway-latency extras only exist in demo. The gateway indicator is now a green pill next to the router name, shown only on the router whose
roleBadgeis Principal (the real gateway), and the whole footer row (label + "View details") was removed because the whole card is already clickable (#593).
Changed
- Spanish UI copy moves from "gateway" to "puerta de enlace" (card pill, "Estado de la puerta de enlace", AP backhaul links/latency, AdGuard/reservation/alert captions, Orchestration copy and the via-Internet tooltip). English text unchanged.
Español
Corregido
- Las tarjetas de "Dispositivos" ya no marcan a todos los routers como "Gateway principal". En modo live el footer caía a esa etiqueta porque los extras de latencia al gateway solo existen en demo. La indicación ahora es una pill verde "Puerta de enlace" junto al nombre, visible solo en la tarjeta del router con
roleBadge Principal(el gateway real), y se elimina toda la fila del footer (etiqueta + "Ver detalle") porque toda la tarjeta ya es clicable (#593).
Cambiado
- Copy en español sin anglicismos: de "gateway" a "puerta de enlace" (pill de tarjeta, "Estado de la puerta de enlace", enlaces/latencia del detalle de AP, captions de AdGuard/reserva/aviso, copy de Orchestration y tooltip vía Internet). El texto en inglés no cambia.
v2.28.7
v2.28.7
Fixed / Arreglado
- OpenWrt
.apkname collision (#573, fix #586): the package workflow built the.apkunder two targets (qualcommax/ipq807x and armsr/armv8) that produce the same filename (no arch in the name), so on upload (--clobber) only one survived. This release now publishes a singleaarch64_generic.apk(armsr/armv8), installable on generic and cortex-a53 aarch64; the cortex-a53.ipkstays for opkg.
Colisión de nombre de los .apk OpenWrt (#573, fix #586): el workflow construía el .apk con dos targets que generaban el mismo nombre (sin arch), así que en el upload solo sobrevivía uno. Ahora se publica un único .apk aarch64_generic, instalable en aarch64 genérico y cortex-a53; el .ipk cortex-a53 se mantiene para opkg.
Changelog completo en CHANGELOG.md.
v2.28.6
v2.28.6
Fixed / Arreglado
- Installers did not parse the release tag (#564):
install.sh,install-agent.shandinstall-collector.shresolvedreleases/latestwith a positionalcut, so when GitHub returns a minified single-line payload the version became the release URL and the download broke. They now extracttag_nameby name, and the installer prints its own version.
Los instaladores no parseaban el tag (#564): resolvían releases/latest con un corte posicional; con la respuesta minificada el tag salía como la URL de la release y la descarga fallaba. Ahora extraen tag_name por nombre e informan de su versión.
- dnsmasq lease file in non-standard locations (#568): now resolved via UCI (
dhcp.@dnsmasq[0].leasefile) with fallback, so hostnames show on routers with a custom lease path. Applies to the agent and the SSH poller.
Fichero de leases de dnsmasq en ubicación no estándar (#568): ahora se resuelve vía UCI con fallback, así que los hostnames aparecen aunque el leasefile esté en una ruta custom.
- SSH host-key mismatch locked the gateway (#567): the discovery probe (system ssh) and the Go pool negotiate a different host key and share
known_hosts; on a known but different key the pool now re-onboards with a warning (accept-new with refresh) instead of rejecting.
Mismatch de host-key bloqueaba el gateway (#567): el pool ahora re-onboarda con aviso en lugar de rechazar.
- Hardcoded timezone and fake 'firmware up to date' (#566): the router info no longer shows
Europe/Madridor an 'up to date' pill before real data; unknown values stay empty.
Timezone hardcodeado y firmware 'actualizado' falso (#566): ya no se muestra Europe/Madrid ni una pill 'actualizado' sin datos reales.
Added / Añadido
- Generic aarch64 apk (#573): the agent and server now publish an
aarch64_genericapk (armsr/armv8), installable on generic and cortex-a53 routers;apk addno longer fails withuninstallable archon non-cortex-a53 platforms.
Apk aarch64 genérico (#573): se publica un apk aarch64_generic instalable en genérico y cortex-a53.
- Honest native vs NetGrip install choice (#570): the agents section explains both options (description, NetGrip port 8090, recommendation) for OpenWrt routers without an agent. NetPulse does not install NetGrip.
Elección honesta agente vs NetGrip (#570): la sección de agentes explica ambas opciones con el puerto 8090.
- Honest 'no per-client traffic' message (#574): the device detail notes wired clients need nlbwmon (WiFi uses hostapd), and the nlbwmon RAM caveat is documented.
Mensaje honesto de 'sin tráfico por cliente' (#574): se documenta el caveat de nlbwmon y la fuente por cliente.
Other / Otros
- #569: rearm a stale NetGrip agent by restarting the netgrip service (merged PR from another session).
#569: rearmado de un NetGrip stale reiniciando el servicio netgrip.
Changelog completo en CHANGELOG.md.
v2.28.5
v2.28.5
Fixed / Arreglado
- Agent copy-command was not runnable (#571): the app generated the install one-liner with space-separated options (
--binary /tmp/netpulse-agent ...), but the installer only parsed--opt=value, so it aborted on the first option withopción desconocida: --binaryand never reached SSH. The installer now accepts both--opt valueand--opt=value.
El "copy command" para instalar un agente no era ejecutable (#571): la app generaba el one-liner con opciones separadas por espacio, pero el instalador solo aceptaba --opt=value, así que abortaba en la primera opción con opción desconocida: --binary y no llegaba al SSH. El instalador ahora acepta ambas formas.
- Installer did not use the server SSH key (#572):
install-agent.shranssh root@IPwithout an explicit identity; when the router only authorized the key NetPulse generates (shown in Settings), SSH failed even thoughssh -i <key> root@IPworked. The installer now accepts--ssh-key <path>, forwards it with-ito every ssh/scp call, and the copy-command includes it automatically.
El instalador no usaba la llave SSH del servidor (#572): install-agent.sh hacía ssh root@IP sin identidad explícita; si el router solo autorizaba la llave que genera NetPulse (la que muestra Ajustes), el SSH fallaba aunque ssh -i <llave> root@IP funcionara. Ahora se acepta --ssh-key <ruta>, se reenvía con -i a todos los ssh/scp, y el copy-command la incluye automáticamente.
Changelog completo en CHANGELOG.md.
v2.28.4
v2.28.4
Added / Añadido
- Proxmox host and container identification via the cluster API (#561): the device detail and topology now seal Proxmox VE hosts as hypervisor and their CTs/VMs as containers attached to the right host, with real names and IPs. The network heuristic (FDB/OUI) could not infer the container→host relationship when a router port mixes devices or CTs use non-standard MACs; the relationship lives in the cluster. The integration is read-only (API token with audit privileges,
PVEAuditor) and is configured in Settings > Proxmox VE (cluster URL + token). Note: in PVE 9.x token privileges are not editable from the UI; assign them via CLI (pvesh set /access/acl -path / -roles PVEAuditor -tokens "user@realm!token").
- Identificación de hosts Proxmox y sus contenedores vía la API del cluster (#561): el detalle de Dispositivos y la topología ahora sellan correctamente los hosts de Proxmox VE como hipervisor y sus CTs/VMs como contenedores colgados de su host, con nombres e IPs reales. La heurística de red (FDB/OUI) no podía deducir la relación contenedor→host cuando un puerto del router mezcla dispositivos o los CTs usan MACs no estándar; la relación vive en el cluster. La integración es de solo lectura (token API con privilegios de auditoría,
PVEAuditor) y se configura en Ajustes > Proxmox VE (URL del cluster + token). Nota: en PVE 9.x los privilegios de un token no se editan desde la UI; se asignan por CLI (pvesh set /access/acl -path / -roles PVEAuditor -tokens "usuario@realm!token").
Changed / Cambiado
- CI: Node 24 (Active LTS Krypton) with pinned
.nvmrc/engines(#542, #560): workflows now run on Node 24 and the repo declares the version for reproducibility.
- CI: Node 24 (Active LTS Krypton) y
.nvmrc/enginesfijados (#542, #560): los workflows usan Node 24 y el repo declara la versión para reproducibilidad.
Full changelog: https://github.com/gnacho/netpulse/blob/v2.28.4/CHANGELOG.md
v2.28.3
v2.28.3
Added / Añadido
- Per-client traffic in live mode (#551): the device detail now shows real per-client traffic (current, total and a chart for 24h / 7d / 30d ranges) when the router's source reports it. Per router the preferred source is nlbwmon (covers wired clients and everything the router routes); when unavailable it falls back to per-station counters from
ubus hostapd get_clients(bytes.rx/tx, wifi clients on that AP), which the agent now reports. The per-client bandwidth store (from #337/#357) is wired end to end: delta ingest with rollups that compute real rates (previously left bps at 0), aGET /api/devices/{mac}/trafficAPI and a live per-client rate in the overview. Note: wired clients behind the gateway will not show traffic until its pusher (NetGrip) emits theclientBwsection; on a router running the NetPulse agent with nlbwmon installed everything is covered.
- Tráfico por cliente en vivo (#551): el detalle de un dispositivo ahora muestra su tráfico real (actual, total y gráfico por rangos 24 h / 7 d / 30 d) cuando la fuente del router lo reporta. Por router la fuente preferente es nlbwmon (cubre cableados y todo lo que enruta); si no está disponible se usan los contadores por estación de
ubus hostapd get_clients(bytes.rx/tx, clientes wifi asociados a ese AP), que el agente ya reporta. La store de ancho de banda por cliente (del #337/#357) queda cableada de extremo a extremo: ingesta por deltas con rollups que calculan tasas reales (antes dejaban los bps a 0), APIGET /api/devices/{mac}/trafficy tasa actual por cliente en el overview. Nota: los clientes cableados del gateway no tendrán tráfico hasta que su pusher (NetGrip) emita la secciónclientBw; en un router con agente NetPulse y nlbwmon instalado queda cubierto todo.
Changed / Cambiado
- Channel analysis redesigned LuCI-style (#542, #545, #557): one chart per band (2.4 GHz and 5 GHz) so the frequency gap and channel label overlap disappear; X axis with uniform channel slots per band; router selector shows all available (those with scans first) and warns when a router reports no scans (NetGrip); signal noise below -90 dBm is ignored.
- Análisis de canales rediseñado al estilo LuCI (#542, #545, #557): una gráfica por banda (2.4 GHz y 5 GHz) para que el hueco de frecuencia y el solape de etiquetas de canal desaparezcan; eje X con slots de canal uniformes por banda; el selector de router muestra todos los disponibles (primero los que escanean) y avisa cuando uno no reporta escaneos (NetGrip); se ignora el ruido por debajo de -90 dBm.
Full changelog: https://github.com/gnacho/netpulse/blob/v2.28.3/CHANGELOG.md
v2.28.2
v2.28.2
Server + frontend. The agent is unchanged (AGENT_VERSION stays 2.26.11).
EN:
- Channel analysis: the X axis now shows every channel of the active band (2.4: 1-13; 5: UNII-1..4), including empty ones, like LuCI; your own managed network is drawn as a violet mountain on its in-use channel at -25 dBm (it is not present in external scans); and one band is shown at a time (2.4 GHz / 5 GHz toggle, no "All") (#542, #554).
- Device actions: the DHCP reservation (GET/PUT/DELETE /api/devices/{mac}/reservation) now targets the router that serves the device's lease, not always the global gateway. On a separate LAN whose DHCP is granted by another router (an AP/lab with its own dnsmasq, e.g. rt-lab on 192.168.2.x), the static host section was being written to the wrong router. When no lease is known it falls back to the gateway, as before (#537).
ES:
- Análisis de canales: el eje X muestra ahora todos los canales de la banda activa (2.4: 1-13; 5: UNII-1..4), incluidos los vacíos, como en LuCI; la red gestionada se dibuja como una montaña violeta en su canal en uso a -25 dBm (no aparece en los escaneos externos); y se muestra una sola banda a la vez (selector 2.4 GHz / 5 GHz, sin "Todos") (#542, #554).
- Acciones de dispositivo: la reserva DHCP (GET/PUT/DELETE /api/devices/{mac}/reservation) ahora apunta al router que sirve el lease del dispositivo, no siempre al gateway global. En una LAN separada cuyo DHCP lo concede otro router (un AP/lab con su propio dnsmasq, p. ej. rt-lab en 192.168.2.x), la sección host estática se escribía en el router equivocado. Cuando no hay lease conocido cae al gateway, como antes (#537).
v2.28.1
v2.28.1
Server-side fix. The agent is unchanged (AGENT_VERSION stays 2.26.11).
EN: SSH polling no longer saturates low-end APs (#550). When the server polls a router over SSH (no agent), it still used the pre-#373 command pair: iwinfo assoclist for clients and an iwinfo radio summary on every 5s tick. On a single-core AP (e.g. Archer A7 v5) that spiked load. The SSH path now mirrors the agent fix from #373: clients via ubus hostapd get_clients (authoritative even on an empty AP, no iwinfo fallback per poll), iwinfo only as a single-pass combined fallback, and the radio summary cached for 5 minutes instead of refreshed every poll.
ES: El sondeo SSH ya no satura APs de bajas prestaciones (#550). Cuando el servidor sondea un router por SSH (sin agente), seguía usando el par de comandos previo al #373: clientes por iwinfo assoclist y resumen de radios por iwinfo en cada tick de 5 s. En un AP de un solo núcleo (p. ej. Archer A7 v5) eso disparaba la carga. La vía SSH ahora replica el fix del agente (#373): clientes vía ubus hostapd get_clients (autoritativo incluso con un AP vacío), iwinfo solo como fallback combinado en una pasada, y el resumen de radios cacheado 5 minutos en lugar de refrescarse en cada poll.
v2.28.0
NetPulse v2.28.0
EN: Scheduled unattended firmware upgrades (#494), wired static-IP hosts discovered via the router ARP table (#507), a bounded poller stop so slow SNMP can no longer hang shutdown (#481), hardened device actions - reserve IP / block device with conflict checks and rollback (#439) - and a full user manual by area and menu, en + es (#418). Note: the 2.26.13/2.26.14 releases were never published (a committed app/node_modules symlink broke goreleaser); their code is consolidated here.
ES: Upgrades de firmware programados y desatendidos (#494), descubrimiento de hosts cableados con IP estática vía tabla ARP (#507), apagado acotado para que un SNMP lento ya no cuelgue el servicio (#481), acciones de dispositivo endurecidas - reservar IP / bloquear con detección de conflictos y rollback (#439) - y manual de usuario completo por área y menú, es + en (#418). Nota: las releases 2.26.13/2.26.14 no llegaron a publicarse (un symlink app/node_modules commiteado rompía goreleaser); su código se consolida aquí.
Changes
- feat(firmware): schedule unattended firmware upgrades (#494)
- feat(devices): discover wired static-IP hosts via router ARP table (#507)
- fix(poller): bound Stop wait so a slow SNMP tick cannot hang shutdown (#481)
- fix(devices): harden device actions with conflict checks, rollback and tests (#439)
- docs: add user manual by area and menu, es/en (#418)
- fix(release): untrack app/node_modules symlink that broke goreleaser
v2.26.12
NetPulse v2.26.12
EN: Channel plan: DFS current-channel score no longer shows a MaxInt value, and the empty state explains why a router shows no scans (#518). Stale failed firmware upgrade notices can now be dismissed (#519).
ES: Plan de canales: el score del canal actual en DFS ya no muestra un valor gigante, y el estado vacío explica por qué un router no muestra escaneos (#518). Los avisos de actualización de firmware fallidos obsoletos ahora se pueden descartar (#519).
Changes
- fix(channelplan): score current DFS channel and clarify empty state (#518)
- feat(firmware): dismiss stale failed upgrade notice (#519)
v2.26.11
NetPulse v2.26.11
EN: Periodic WAN speed test against speedtest.net (#511), fix for the false "update failed" (update_exit_-1) with human-readable update errors (#512), and a RAM usage metric that no longer counts reclaimable cache (#513).
ES: Test de velocidad WAN periódico contra speedtest.net (#511), arreglo del falso "la actualización falló" (update_exit_-1) con errores legibles (#512), y métrica de RAM que ya no cuenta la caché reclamable como consumo (#513).
Changes
- feat(speedtest): periodic WAN speed test for connection health (#511)
- fix(updater): match success marker by prefix and surface human-readable update errors (#512)
- fix(agent): report RAM usage excluding reclaimable cache (#513)
v2.26.10
EN
Wired clients behind non-standard bridge ports are now discovered (#506). The agent filtered bridge FDB MACs through a hardcoded port-name allowlist; physical ports with unusual names (the BPI-R4 SFP cages are named sfp-lan/sfp-wan on OpenWrt 25.12+) were silently dropped, so static-IP hosts behind them (e.g. behind an unmanaged switch, no DHCP lease) never showed up as clients. The filter is now a denylist (wireless and virtual interfaces excluded). Also fixed: the bridge fdb show path no longer leaks local entries, which could make the server discard all wired clients of a port. Embedded agents now report 2.26.10, so every older fleet agent gets offered the update.
ES
Los clientes cableados tras puertos con nombres no estándar ya se descubren (#506). El agente filtraba las MACs del FDB del bridge con una lista cerrada de nombres de puerto; los puertos físicos con nombres inusuales (las jaulas SFP del BPI-R4 se llaman sfp-lan/sfp-wan en OpenWrt 25.12+) se descartaban en silencio, y los hosts con IP estática tras ellos (p. ej. detrás de un switch no gestionado, sin lease DHCP) nunca aparecían como clientes. El filtro pasa a ser una lista de exclusión (fuera wireless e interfaces virtuales). También corregido: la vía bridge fdb show ya no deja pasar entradas locales, que podían hacer que el servidor descartase todos los clientes cableados de un puerto. Los agentes embebidos ahora reportan 2.26.10, así que a toda la flota anterior se le ofrece la actualización.
v2.26.9
Fixed
- The in-app help section now spans the same width and text size as the rest of the app (#503): it was self-constrained to 768px with 13px body text; it now follows the standard container (1160px at a 1440px viewport, same as Home) and the app's text-sm body size.
- La sección de ayuda in-app ahora ocupa el mismo ancho y tamaño de letra que el resto de la app (#503): se autolimitaba a 768px con cuerpo de 13px; ahora sigue el contenedor estándar (1160px a 1440px de viewport, igual que Inicio) y el text-sm de la aplicación.
v2.26.8
Added
Apply the recommended channel from the channel plan page (#500)
The agent now reports each radio's UCI section and the plan page becomes actionable: per-radio Apply button with confirmation (current to target channel, network restart warning), live progress and result feedback, and a one-click revert to the previous channel in the same session. Plans go through the existing plan engine (explicit diff) via the NetGrip executor or agent SSE. Full ES/EN translation of the page and current/best score per radio for context.
El agente ahora reporta la sección UCI de cada radio y la página del plan pasa a ser accionable: botón de aplicar por radio con confirmación (canal actual a objetivo y aviso del reinicio de red), progreso y resultado en vivo, y vuelta al canal anterior con un clic en la misma sesión. Los planes van por el motor existente (diff explícito) vía executor de NetGrip o SSE al agente. Traducción completa ES/EN de la página y puntuación actual/mejor por radio como contexto.
Fixed
- POST /api/plans with an explicit diff returned 201 with an empty body (empty desired from the DB failed to encode, silently swallowed); JSON clients got EOF.
- The apply-result report died when the ops restarted the router's own network, leaving plans stuck in applying; the agent now retries with backoff for several minutes.
- POST /api/plans con diff explícito devolvía 201 con body vacío (el desired vacío de la BD fallaba al codificar y el error se silenciaba); los clientes JSON recibían EOF.
- El reporte del apply moría cuando las ops reiniciaban la red del propio router, dejando planes en applying eterno; el agente ahora reintenta con backoff durante varios minutos.
v2.26.7
What's new / Novedades
EN — Two quality-of-life fixes from the forum feedback wave. The CSP now allows inline styles, so Radix dialogs (command palette included) stop logging a console violation on every open and their enter/exit transitions actually run; script-src remains strictly 'self'. And there is a new in-app Help section with guided walkthroughs for the five flows a first-day admin needs — first login and password, onboarding a router, installing an agent (one-click button first, pairing-token one-liner as the manual fallback), reading the channel plan, and running a firmware upgrade — in Spanish and English, quoting the real UI labels as chips so the help drifts together with the interface instead of going stale.
ES — Dos arreglos de calidad de vida de la tanda de feedback del foro. El CSP ahora permite estilos inline, así que los diálogos de Radix (paleta de comandos incluida) dejan de loguear una violación de consola en cada apertura y sus transiciones de entrada/salida por fin se ejecutan; script-src sigue estricto. Y hay una nueva sección de Ayuda dentro de la app con recorridos guiados para los cinco flujos que un admin necesita el primer día — primer arranque y contraseña, alta de un router, instalación del agente (botón de un clic primero y one-liner con token de pairing como alternativa manual), lectura del plan de canales y upgrade de firmware — en español e inglés, citando las etiquetas reales de la UI como chips para que la ayuda derive junto con la interfaz en vez de quedarse vieja.
Closes #485
Closes #484
v2.26.6
Changed
Safer firmware upgrades page (#477)
The list only includes OpenWrt/GL.iNet routers with an agent (managed switches and beacon-only devices are gone from the page), the form prefills model and target detected by the agent itself, and starting an upgrade requires confirming the target (current to target version, image URL, sha256 verification, reboot warning). Stale upgrade errors on the card carry a relative timestamp so an old failure no longer reads as a dead agent. Scheduled upgrades tracked separately in #494.
La lista solo incluye routers OpenWrt/GL.iNet con agente (los switches gestionados y los dispositivos solo-beacon desaparecen de la página), el formulario se prefiere con el modelo y target detectados por el propio agente y arrancar un upgrade exige confirmar el objetivo (versión actual a objetivo, URL de la imagen, verificación sha256 y aviso de reinicio). Los errores de upgrade antiguos de la tarjeta llevan marca de tiempo relativa para que un fallo viejo no se lea como agente caído. Las actualizaciones programadas quedan como seguimiento en #494.
v2.26.5
What's new / Novedades
EN — The agent now collects LLDP neighbours itself: every full probe cycle it reads lldpcli and ships the neighbours (local port, chassis, management address, capabilities, remote port) in its push. Until now LLDP was only collected by the server over SSH, a path that routers with an agent never take, so the topology map could never promote managed switches or identify LLDP-speaking devices on those fleets. The payload contract is explicit: available:false when lldpd is not installed (feeds the in-app hint), missing neighbours when the probe fails that round (anti-flicker), and an honest empty list when there is simply nobody to hear. Network note: a conformant 802.1D switch filters LLDP multicast, so neighbours show up between directly cabled equipment or across unmanaged switches.
ES — El agente ahora recolecta los vecinos LLDP por sí mismo: en cada ciclo completo de sondeo lee lldpcli y los envía en su push (puerto local, chasis, dirección de gestión, capacidades, puerto remoto). Hasta ahora el LLDP solo lo recogía el server por SSH, una vía que los routers con agente nunca usan, así que el mapa de topología no podía promover switches gestionados ni identificar dispositivos que hablan LLDP en esas flotas. El contrato del payload es explícito: available:false si lldpd no está instalado (alimenta el hint de la app), vecinos ausentes si la sonda falla esa ronda (anti-parpadeo) y lista vacía honesta cuando no hay nadie que escuchar. Nota de red: un switch conforme 802.1D filtra el multicast LLDP, así que los vecinos aparecen entre equipos directamente cableados o tras switches no gestionados.
Closes #489
v2.26.4
What's new / Novedades
EN — The agent now ships for MIPS devices: mipsle soft-float (MT7621 and the rest of the little-endian targets, the most common case) and mips soft-float (big-endian, ath79 and friends). Available as release tarballs, through the in-app install/reinstall buttons (the server embeds both binaries and serves the right one), and via the one-liner, which now detects the endianness on its own since uname -m just says "mips" for both. If you have an MT7621 box that was left out, this release is for it.
ES — El agente ya se compila para dispositivos MIPS: mipsle soft-float (MT7621 y demás targets little-endian, el caso más común) y mips soft-float (big-endian, ath79 y compañía). Disponible como tarballs de release, a través de los botones de instalar/reinstalar de la app (el server embebe ambos binarios y sirve el correcto) y con el instalador de una línea, que ahora detecta el endianness por sí mismo porque uname -m dice "mips" para ambos. Si tienes un equipo MT7621 que se quedaba fuera, esta release es para él.
Closes #488
v2.26.3
What's new / Novedades
EN — The update wizard now shows a real changelog before you apply an update: the list of commits between your version and the latest one (short sha + subject, newest first), how many changes are coming, and a link to the full comparison on GitHub. Works in both rolling (SHA) and stable (tag) modes, caches the result so the daily check does not repeat it, and degrades gracefully: if the comparison cannot be fetched, the wizard falls back to the previous release-notes view.
ES — El asistente de actualización ahora muestra un changelog de verdad antes de aplicar: la lista de commits entre tu versión y la última (sha corto + asunto, lo más reciente primero), cuántos cambios entran y un enlace a la comparación completa en GitHub. Funciona tanto en modo rolling (por SHA) como estable (por tag), cachea el resultado para no repetir la consulta en cada chequeo diario y degrada sin romper nada: si el compare no se puede consultar, el asistente vuelve a la vista de notas anterior.
Closes #490
v2.26.2
Changelog
- 45d98f5740893aa3c38b248237ea166b0d3ba5f3 chore(release): v2.26.2
v2.26.1
Changelog
- 9364340ded568ed948a4f0f770316be7c922a42f chore(release): v2.26.1
v2.26.0
What's Changed
- feat(v2.26): Fases 2-4 - UCI ownership, channel plan y firmware upgrades by @gnacho in https://github.com/gnacho/netpulse/pull/456 (closes #451, closes #452, closes #453)
- fix(ci): build agent binaries before tests and name arm output correctly by @gnacho in https://github.com/gnacho/netpulse/pull/459 (closes #458)
Detalles
- Apply seguro de configuración (#451): ownership UCI por integración y rollback automático si el health check post-apply falla.
- Planificación de canales (#452): escaneos pasivos de vecinos desde el agente y recomendaciones de canal por radio.
- Firmware upgrades (#453): targets de firmware con URL + SHA256, comando SSE al agente, descarga verificada, backup de config, sysupgrade y reporte post-reinicio. Validado E2E en router real con imagen oficial ASU.
- Watchdog:
pidofen lugar depgrep -x(BusyBox no matcheaba y reiniciaba el agente cada 2 min). - CI: binarios de agente antes de los tests y nombre
netpulse-agent-armcorrecto.
Full Changelog: https://github.com/gnacho/netpulse/compare/v2.25.0...v2.26.0
v2.25.0
What's Changed
- feat(devices): DHCP lease column + device alias/icon overrides (#437) by @gnacho in https://github.com/gnacho/netpulse/pull/438
- Hide vitals for sources without system metrics, open web UI/NetGrip actions, reinstall in Agents section by @gnacho in https://github.com/gnacho/netpulse/pull/445
- fix(ui,server): timeline compacta de upgrades y comparación de builds (#446, #447, #449) by @gnacho in https://github.com/gnacho/netpulse/pull/450
Full Changelog: https://github.com/gnacho/netpulse/compare/v2.24.0...v2.25.0
v2.24.0
Novedades
- Asesor de re-anchor WiFi (#403): nuevo panel "Reanclar" en
/roamingque sugiere mover clientes WiFi a un AP con mejor señal. Soporta usteer (preferido) y DAWN como fallback.
Cambios técnicos
server-go/internal/adapters/reanchor.go: motor de recomendaciones conubus call usteer get_clients(hearing map).server-go/internal/httpapi/reanchor.go: endpoints/api/wifi-reanchor/recommendationsy/api/wifi-reanchor/{mac}/move.app/src/pages/ReanchorPanel.tsx: UI de recomendaciones y acción manual.- Tests unitarios y bump de versión a 2.24.0.
v2.23.5
Changelog
- 3d7bfa9166eff31dc4499fa00fadd96143ac30cc chore(release): v2.23.5
v2.23.4
Changelog
- f523d2095a1d804d633eb6cec0f29fee5acaf97f chore(release): v2.23.4
v2.23.3
Changelog
- f2a48664bc73da516bd8015dc01ea521b65545a2 chore(release): v2.23.3
v2.23.1
Fixed\n\n- Regresion en modo glinet de AdGuard: tras #421, PUT /api/config/adguard rechazaba host vacio en modo glinet. Ahora se acepta de nuevo y se devuelve port: 0. (#423)\n\nActualiza desde Ajustes > Actualizaciones en la app.
v2.23.0
Changelog
- d6ae3ca8cb64a89b2afb95d5292e31bee70b0afe fix: AdGuard port persistence and client IP ARP fallback (#420, #421)
v2.22.2
Changelog
- f0fd9d67c25e864dd837c6fc9cebdb92e8677ea6 chore(release): v2.22.2
v2.22.1
Changelog
- 22e5d2e7077c84386a0eb534c075918d5dabcc38 chore(release): v2.22.1
v2.22.0-2
Changelog
- 65641912c9a7669bd510eaafad4ca313569f12b4 chore: add DAWN test key fingerprints to .gitleaksignore
v2.22.0
Changelog
- e5f8d168f74ae390a2c907ec4afe763b9bb95b1c chore(release): v2.22.0
v2.21.0
Changelog
- e19fe67e8ba0ec8fbf2748ea6e9214b3ece964f8 chore(release): v2.21.0 (#399)
v2.20.0
Changelog
- 557ba18acb12aa3e329e717f8c50327c92c7f277 chore(release): v2.20.0 (#396)
v2.19.2
Seguridad
- Deteccion de placeholders: rechazo de valores del .env.example en AUTH_PASS y SESSION_SECRET al arrancar (fail-fast, #390)
Leccion de la auditoria de Yuvomi v2.54.0 (issue upstream #937).
Otros patrones de Yuvomi ya cubiertos en NetPulse:
- Anti-enumeracion: dummyPasswordHash en auth.go (timing-safe)
- Skeleton/EmptyState: componentes propios en uso
v2.19.1
Changelog
- 651cbddef58af855711d03584cb70bbf3d4b3b62 chore(release): v2.19.1
v2.19.0
Changelog
- ea8caea5abd6e2534fd1494a3662ff8c80386c7a fix: pathanalysis Summaries deadlocks with MaxOpenConns(1) (#374)
v2.18.0
Changelog
- e9979ebb8ef19599478947a8eeb2c65765fa041b release: v2.18.0 (#324)
v2.17.0
Changelog
- c6a4724d42ba1818b53bd3a709af10eb0004a111 release: v2.17.0 (#322)
v2.16.0
Changelog
- aa1fba060b325983b98cd64cbdd4f774a4c447dd release: v2.16.0 (#320)
v2.15.0
Changelog
- 075277688ad8cf79d97fd2b476fd0fb74aced89a release: v2.15.0 (#296)
v2.14.0
Changelog
- edfce643985fd5ae61278ad880267d62045f1c46 release: v2.14.0 (#287)
v2.13.1
Changelog
- b8afccf4cb5b7efb9401f72e6b9c6c03c8e47e5f fix(ui): keep router back button fixed at top as icon-only (#278)
v2.13.0
Changelog
- d5d860488e1f4c89afc3495a19a7e5553960527e release: v2.13.0 — reorganized Resumen hero card (#267)
v2.12.0
Changelog
- 19e8979ca32cf57bb57794fa7055f73a4d94e628 release: v2.12.0 — backlog sweep (backend fixes, frontend, demo, WireGuard orchestration) (#264)
v2.11.2
Changelog
- 6ee6c5c5897984aea8a5d4e41d2f59b68aef4bd8 release: v2.11.2 — backlog sweep (security, topology, frontend, demo, packaging)
v2.11.1-2
Changelog
- f4300f27f2ffbdccfd4def6148bdbe6dc7e6e064 ci: report the release version in the server health endpoint
v2.11.1
Changelog
- 074249447bc10c1790fb7e84b6735f552eca6906 fix(agents): deadlock in GET /api/agents when routers have a type
v2.11.0
Changelog
- b3daedb803e63b833300ada696fb05155cbc8bb0 release: v2.11.0 — audit round (firmware target, SSH rotation, agent upgrade & reinstall) (#241 #242 #243 #246) (#250)
v2.10.2
Changelog
- 181331f82d9e8051e695c8e338c3876b7c5d232b release: v2.10.2 — trusted devices allowlist and unknown-device alert fix (#196) (#198)
v2.10.1
Changelog
- 96a0705258024e78ebad896a27e1f3abd420cbad release: v2.10.1 — fix update banner duplicate version and stable hint (#191) (#193)
v2.10.0
Changelog
- b341732ffb05af058372a241f6c3961132fd796b release: v2.10.0 — device offline/online events (#184) (#186)
v2.9.6
Changelog
- 61c2e1f2cf9baa2b76a380cee72b9b6b9d81644a fix(update): framer-motion exit accepts undefined not false (#182)
v2.9.5
Changelog
- 2a72b37a7e30242f04dd652b21104d7fc370fcd4 release: v2.9.5 — ingesta benigna en demo (#177)
v2.9.4
Changelog
- 7b51b4bd37fb7845f60a925934dadf6d716a4151 release: v2.9.4 — fix móvil Home (#175)
v2.9.3
Changelog
- cb3893d05e8e63e338f66b8012616347db1891e5 release: v2.9.3 — WAN summary stats from metrics in live mode (#171)
v2.9.2
Changelog
- cf64d46863e7e031cdf5354d16cbae60d8a67216 fix(settings): remove unused reduce prop from ExternalDevicesManager
v2.9.1
Changelog
- 5a9e095dfdf706e5f1f64a4b9076a818c70ac3e1 release: v2.9.1 — orchestration 17.2-17.4, portable topology, About links (#164)
v2.9.0
Changelog
- 94858469209a748fbbe52addb5e4eed849b847c4 release: v2.9.0 — Fase 17.1 AdGuard + fixes #107 #108 (#125)
v2.8.0
Changelog
- a97ea7cc0e7c9477cd5b0b0d3f38bf77ec4f5b76 release: v2.8.0 — Phase 14 complete (Survey + Events) + router UPDATE
v2.7.2
Changelog
- 8d2a71080094d5681d5021f871e13b77a3028a8e release: v2.7.2 — auditoría Fable (single-flight, SSE deadline, sshpool race, error wrapping)
v2.7.1
Changelog
- fd5c0b7e82a03463944c74bc19c5a036265780fa feat(version): v2.7.1 — changelog y bump de versión (fix panic poller #60)
v2.7.0
Changelog
- 11217db2c9a5154d310a6d3eefeb99abaeddf026 chore(app): Tailwind 4 (CSS-first @theme) + react-is explícito + tw-animate-css (#35)
- 4fc938ade6b61ba8f65329f1ca3c7893e8910c31 chore(app): TypeScript 7 (tsgo) — quitar baseUrl (eliminado en TS7) (#34)
- 2b3c8cf88da1d61c31928e213f39711c0e42698f chore(app): Vite 8 + plugin-react 5.2 (requiere Tailwind 4) (#36)
- 165f4aa3460d0b9b0b7fab4394a476d0918f4a09 chore(app): react-router 8.3.0 (react-router-dom fusionado en react-router) (#37)
- e1361b4d17b5885f7ba82f45a2b2d642c8532c93 chore: bump modernc.org/sqlite a v1.56.0 (server-go + collector) (#33)
- d93c923964c39f224273d4aba4a2861f7d5bd5c0 docs(roadmap): Fase 10 paquete luci-app-netpulse
- d809b4ef0825558840b9a2eb835d947260c583d8 docs(roadmap): Fase 7 completada, estado actual v2.6.0
- dd53f6917bde1a4ccf944024ae2d1d73c1d0ed59 docs(roadmap): Fase 8 completa — marcar ítems HECHOS con PR, reordenar, limpiar líneas obsoletas
- 8936131c893f2717ec4bf2ddda5fbd98b403fdf4 docs(roadmap): subir todo el stack frontend (TS7, Vite8, Tailwind4, react-router8) — revierte decisión 8.6
- 4a4428e8ed81a0aa1ce7b0764828b127c3edda27 docs(roadmap): webhooks y API de ingesta en Fase 9
- 793b5722dd4172c7a9653a8fdc51f06f1fdc612f docs(roadmap): ítem 3 — retención larga en server-go (descartada la frontera sidecar)
- ab2049a2e2ded0dbf79f4098a9c5ae08c96d8200 docs(roadmap): ítem 7 — saliente prioritario, entrante diferido (skill api-ingest-go creada)
- f5be84ae016be7849c31bdb34fe2a51d9feb14ff docs(roadmap): ítem 8 completo — lecciones del bloque de actualización
- 3ba76c13d65c9cdca1eae5cc666f7abc9d393496 docs(roadmap): ítem 8 — actualización del stack con seguimiento por componente
- 439722778e1bf3fccd6cc94af4d0a46488687436 feat(app): recharts v3 (2.15.4→3.10.1) y eliminar wrapper chart.tsx sin uso (Fase 8.4) (#29)
- 18bf47b03b70a0eab07f597b88b8001268b88ae3 feat(ci): workflow de empaquetado OpenWrt (.ipk + .apk)
- 5bfe86682bd9e590a52ecb34f8188042dea5eac6 feat(landing): enlace Ko-fi real (ko-fi.com/gnacho) en about y footer
- 12158b51d25fff1b03859e382d3ff75b54f5e797 feat(landing): lightbox clicable en el slider de capturas + webp regenerados q80 (#41)
- cc2739ba866b6404637977ef09276fd123a49261 feat(landing): slider de capturas por menú + fixes + SEO
- 9af95ffeaf071559847ce0511a967e597a97f941 feat(server): informe semanal — GET /api/reports/weekly (semana ISO) (#50)
- b54c21b1f923966e2f1fc50dbaeb3477aa94e891 feat(server-go): escalera de retención (buckets 5min 1año + daily ∞) y 30d desde buckets (Fase 8.3) (#24)
- 27222c7aa13bdc075e69d9d284c8ec9db26b37e7 feat(server-go): métricas operativas en /api/health (agentsConnected, sseConnections, devicesTotal) (#20)
- 0133f40433c0fa3dac71347419dc849c874f21ad feat(server-go): persistir el registry de agentes en kv (Fase 8.2, R8) (#22)
- e99338c3418ce8d2b2a00a245119ae797f161f70 feat(server-go): webhook saliente de alertas con HMAC, retry y DLQ (Fase 8.7b) (#31)
- 39ec5aa3bfa4aaf41b8b90288f027dd1df94439c feat(settings): AdminBar canónica — Actualizaciones→Usuarios→Modo demo
- 443cc4bedc941f353d672dbf4f8473adeebde234 feat(web): bloque Club Cloudless + Acerca de mí en la landing (#51)
- 16cc4d95569e0b21df0813d4ea98a0b264bc6917 feat(web): landing page muestra NetPulse como app OpenWrt nativa con .ipk
- a8c81b3ca1e1adc4efbc2bea2be53ac4377f7e2d feat(web): vista Informes — disponibilidad semanal por router (#52) (#53)
- 7d186829b337cc3d99710ab1cae4d3631148937c fix(app): lint con TS7 — side-by-side TS6.0.2 + limpieza 40 problemas + job CI (#47)
- 573ea34732065143edddfd8d828811e08fb6d4d8 fix(app): overrides typescript-eslint para que npm ci funcione con TS7 (deuda lint documentada)
- a7ee5246a13e4aca0fb995dc9093dd995b5deb27 fix(collector): inyectar Version real con -ldflags (reportaba 0.1.0) (#48)
- 9347028a7b9840961aab640d10f9c1a732a01d7c fix(landing): captura de topología esperando al fin de la coreografía de montaje (#46)
- 8ddef17cc5b4ab37c77b6c93de7e958d28b6b3f5 fix(landing): capturas a escala real (viewport 1440x900 DPR2 en vez de 2880CSS) (#44)
- 4b2611473a0c8f80d83c81aca38f4ab2e536dbdd fix(landing): re-aplicar hreflang + FAQPage JSON-LD en web/index.html (no habían persistido)
- ab584ffcff6ef79c239ee897735384fe58dc47ee fix(security): auditoría de seguridad — TRUST_PROXY, anti-replay, body cap y password mínima 10 (#59)
- 9cd1f42d95c0f0117f2ea9e7c490d80289d77409 fix(server): rollup nocturno — up_min en minutos + flag NETPULSE_ROLLUP para reparar (#54) (#55)
- f36b283f2cf664f4f07d19d2a065939064d6063a fix(server-go): bump golang-jwt/jwt/v5 a v5.3.1 (CVE-2025-30204) (#16)
- 34a45bdd168d7f3ff728016abb87043cd18f85c3 fix(web): add missing st7/f9/package i18n keys from landing .ipk update (#27)
- bf6b78677c3a5d91760dc3436af1de6b711e6e58 fix(web): bloque Club Cloudless — CTA 'Conocer el club' junto al texto (fullHD) (#56)
- aaf787e9687265e05aea2c791b03d60d70e9c3b9 fix(web): cache-busting on static assets (v query) so landing updates propagate (#28)
- 57e6f9c37e15051598f8fa0921bdfb3fbe3daeb7 fix: --version flag en agente, BuildCommit en servidor, goreleaser actualizado
- c5696957133fd4286005585030aeb87b15c0a6ab refactor: move legacy Node backend to legacy/server-node (history only)
v2.6.0
Changelog
- e1d733fc0ea354257dea244908f6d3b0c72d4cd3 chore: bump v2.6.0
v2.5.0
Changelog
- 5f436cbfaed781ead53700a7cfea130b495051fc fix(agent): eliminar código muerto hasAgentToken y manejar error de ReadSeeker
v2.4.3
Changelog
- d36472cb20777c9563ec89d77f3184b2a39d9f47 fix: gl-clients también por la ruta de agente (issue #5 bug 1)
v2.4.2
Changelog
- b71261ae4888c2b508fd81131e447d5a59260544 Merge pull request #9 from gnacho/fix/issues-3-4-5
v2.4.1
Changelog
- fc48df1479d07701fe41d136a1c838e900871f90 fix(auth): requerir rol admin en rutas de agentes, rearme y gestión de routers (#8)
v2.4.0
Changelog
- 55e552301a4e040087ae95d15181e53f886595fc feat(rearmer): auto-rearme de agentes tras TTL (NETPULSE_AUTO_REARM)
v2.3.0
Changelog
- c5142a42fcc5983671a4f463d6d5fd7f31475cbc test(health): referenciar httpapi.Version en vez de hardcodear 2.2.0
v2.2.0
Changelog
- ee0b1c2e7ddb3e0cc1c2ba623478c1e6512fd11c chore(security): purgar clave SSH de sandbox del historial (filter-repo) + fingerprint push.go en gitleaksignore
v2.1.0
Changelog
- ef7d036eade6e4933b032024a97a9623d13f754f fix(ci): dist embebido deja de trackearse — el placeholder rompía goreleaser --clean (árbol sucio / tag≠HEAD)
v2.0.0
Changelog
- 2565b1a07d6594942dcbde88cb7e4379c591fa40 ci: fix release args — empty string is falsy in GH expressions (tag runs were snapshots)