0.1.0-alpha.5.5

Improvements and additions

• The site is kept out of releases: not in an appliance's download, not in the changelog (#678)
• The TCP window of a connection in hardware, checked there (#679) (#682)
• An appliance carries what it reads: the documentation, the workflows and the repository's tools left out of its archive (#683)
• The heartbeat with the fast path: the XGS 3300's answers every time, the XGS 136's does not (discovery 11) (#687)
• When the fast path stops: reported, shown, and the appliance restarted if its owner chose that (#688)
• The XGS 136's heartbeat: the answer is the beat taken off the host's queue (#689)
• The XGS 3300's coprocessor reached over its PCIe management link instead of its console (discovery 28) (#690)
• Nothing leaves an appliance unless its owner chooses: reports manual and the count off by default, asked once on XGS: Overview - an appliance updated to this release is turned the same way and asked again; and Suricata's inline IPS and the offload said to be, for now, one or the other (#691)

Documentation

• The discoveries of 2026-10-22: what the hardware can do that was not known, and the completion figures lowered to count it (#681)
• The bench photograph as the measurements' main picture, and a news post on what the hardware can do (#684)

0.1.0-alpha.5.4

Improvements and additions

• xgs: the coprocessor pages' Arabic is in opnsense-arabic's catalogue; nothing is merged from here (#674)

0.1.0-alpha.5.3

Improvements and additions

• npuctl: trust the management link at boot when its ssh is refused; a reinstalled host replaces its old key (#668) (#671)
• xgs: a restored configuration's driver settings reach the driver at the first boot (#669) (#672)

Documentation

• docs: an image from the builder on the XGS 136 - installed, updated from the GUI, its configuration restored (#670)

0.1.0-alpha.5.2

Improvements and additions

• site: the add-your-appliance form on the home page too, beside every model, in every language (#665)
• installer: say last when the next boot will have no driver; remove old kernel sources once the new kernel has settled (#666)

0.1.0-alpha.5.1

Improvements and additions

• site: a page for every model and every expansion module, drawn; the menus, the footer and the home page; where the project stands and its releases; what search engines read (#659)
• site: German, Spanish, French, Turkish, Russian and Japanese - the home page, the models, the modules, the status, the releases and how the measurements are taken (#660)
• site: a link in a drawing no longer stops the page from changing; every language in the language menu; favicon.ico and a touch icon (#662)
• reporter: a driver that is not attached is reported with what became of its module - installed or not, built for which kernel, loaded or not - and the last lines of the installer's build of it (#663)

Documentation

• docs: questions people ask, a glossary, how the measurements are taken, and the models' sources in the provenance (#658)

0.1.0-alpha.5

A gathered release. On both families, the appliance's own traffic through a tunnel that strongSwan has
moved onto a narrower path no longer loops in the kernel. On the XGS 3300, a receive ring whose interrupt
was raised and never handled is found within two ticks instead of up to 50 ms, every ring's interrupt has
a CPU of its own, and an arriving ESN association is decrypted by the engine. Every model of both families
now installs, its front ports named from one table by the assembly number on its board; every model but
the XGS 136 and the XGS 3300 is added untested - its status page says so, and its first boot checks its
ports against what its coprocessor reports. The drivers changed: the Updates page asks for a reboot.

Fixes

• npuep, octep: the appliance's own packet too big for the tunnel's path is fragmented, not answered EMSGSIZE - which looped its TCP in the kernel (#647)
• octep: a ring left with a burst - the interrupt raised and never handled; the look each tick, two faults fixed (#651)

Improvements and additions

• xgs: an update asks for a reboot only when the driver the boot loaded changed - not when its file is merely newer (#642)
• octep: an arriving ESN association on the XGS 3300's engine, by default - measured to hold (#645)
• xgs, npuep, octep: every model's front ports from one table, by the assembly number on its board (#652)
• xgs, npuctl: every model installs, and every model not run here is marked research and development; every model as a table (#654)
• xgs, npuep: each port's path from the vendor's device trees, the 138's tags by its coprocessor's order, a port check reported from untested models, tables marked added untested (#655)

Documentation

• README: the two address-drop badges say the same thing (#643)
• docs: the XGS 136's receive marked done on the status picture, and the picture drawn again (#646)
• docs: RESEND does not wake a ring that lost its interrupt (#653)

0.1.0-alpha.4.3

A fix for the XGS 136 that 0.1.0-alpha.4.2 made urgent: reading the switch's port counters every
five minutes - installed everywhere by that release - made the 136's coprocessor report linked
ports down, sometimes for as long as no cable moved (#639). The 136 no longer reads them; after the
update, a reboot of the 136 clears any port it still believes down. And on the XGS 3300, XGS: Ports
counts F1, F2 and the expansion module's ports from the coprocessor's own MACs. No driver changed.

Fixes

• xgs: the XGS 136's switch counters not read - the reads made the network agent report ports down (#640)

Improvements and additions

• xgs: the XGS 3300's own MACs counted on XGS: Ports - F1, F2 and the module's ports (#637)
• website: the release notice's notes formatted, each pull request's number its link, the unsubscribe link plain (#638)

0.1.0-alpha.4.2

Checking for updates no longer runs out of GitHub's hourly allowance: XGS: Updates and its Plugins
tab ask npuforge.org's copy of GitHub's answers first, and what is installed is still confirmed with
GitHub itself. The switch's port counters on XGS: Ports work - their script was never installed - a
first install fetches the kernel's sources it needs to build the driver, and an appliance is counted
on npuforge.org as soon as it is installed or updated. On the XGS 136 the receive runs on the
coprocessor's doorbells by default, with a setting to turn them off; the driver changed, so it takes
effect at the next boot.

Fixes

• xgs: the problem reports of 2026-10-20 - portstats.py installed, one report for one failing action, a first install fetches the kernel's sources (#632)
• xgs: an appliance counted at the end of every install, not the next morning (#635)

Improvements and additions

• xgs: update checks ask npuforge.org's copy of GitHub's API first (#631)
• npuep: the receive on the coprocessor's doorbells by default, and a setting for it (#633)

0.1.0-alpha.4.1

Three things on the pages of 0.1.0-alpha.4, found on the first appliance it was installed on: the
dates at the top of XGS: Updates read "Invalid Date" (since 0.1.0-alpha.3), its Channel menu cut its
text off, and XGS: Ports' switch counters showed empty headings until their first reading. And XGS:
About has a Releases tab: every release's changes, the one installed here marked. Nothing in the
drivers changed; no reboot.

Fixes

• xgs: a Releases tab on XGS: About; the dates on XGS: Updates read right again (#621)
• xgs: XGS: Updates' channel in OPNsense's own picker (#622)
• xgs: the switch's counters read when the page opens, or 'Not available on this appliance' (#624)

Documentation

• website: /stats/ as figures, a chart, the most read pages by name, and the appliances (#619)

0.1.0-alpha.4

A reload of the XGS 136's driver could stop the host when the coprocessor held the tunnel's
encrypting association at that moment (#603); the association now gets the kernel's own ESP transform
back first. XGS: Ports shows the switch's own counters for each port behind it - the errors and drops
the host never sees. The XGS 136 can take its receive from the coprocessor's doorbells
(hw.npuep.giu.rx_msix, off by default until a soak with it on), and a reload of its driver applies
what a boot applies. The XGS 136's driver changed, so it asks for a reboot; the switch tool on the
coprocessor is replaced at the next boot.

Fixes

• npuipsec: the kernel's ESP transform back when the engine lets an association go (#615)

Improvements and additions

• npuep: a reload applies what a boot applies once the ports are there (#607)
• xgs: the switch's own counters for each port on XGS: Ports (#614)
• npuep: the XGS 136's receive rings on the coprocessor's doorbells (#616)
• relay, site worker: the first CodeQL alerts (#612)

Documentation

• website: a link to the Sponsors page, in both languages (#608)
• news: sponsoring the project (#610)
• CITATION.cff: GitHub's Cite this repository (#611)
• docs: two hours of TLS through the XGS 136's IPsec engine (#613)
• docs: six IPsec tunnels at once through the XGS 136 - 96 % (#617)