ircbits.com

Index / IRCv3 now flags AI-assisted clients in its support tables

IRCv3 now flags AI-assisted clients in its support tables

Updated August 9, 2026

The IRCv3 software support tables — the closest thing IRC has to an official capability matrix — now mark software whose development was substantially AI-assisted. The change merged on 4 August 2026 and is live.

It is deliberately quiet: a grey fa-robot icon beside the entry, tooltip "Development assisted using AI", sitting in the same row as the existing operating-system icons. In the underlying data it is a single ai: true flag per entry.

Three clients carry it today:

The Lounge is the one that will surprise people. It is a long-established project, it is what a large share of self-hosters run, and it now sits in the tables with a robot next to its name.

It is also the mark with the least explanation behind it. The original change came with a rationale and named ObsidianIRC specifically. The Lounge's flag arrived three days later, in a pull request whose description was about removing six unmaintained clients, with no accompanying note about what the assessment was based on. We have not found any public justification for it, and The Lounge does not appear to have been asked.

The proposal came from Sadie Powell, InspIRCd's lead developer, and the reasoning in the pull request was about disclosure rather than judgement — marking software "so people reading the software lists can make a fully informed choice". She also argued for a deliberately high bar: mark anything substantially AI-assisted, "to avoid potential witch hunts about whether minor patches are AI or not".

That threshold is doing a lot of work, and it is not yet written down anywhere as a rule. Nor is there a declared process for a project to contest a mark or request one. For now this is a maintainer applying a flag by hand to software they happen to know about — which cuts both ways. An absence of a robot is not a clean bill of health, and a robot is not a finding of fact.

The timing is not coincidental. On 22 July, Codeberg's members voted to change the platform's terms of use to prohibit "vibe-coded" projects — those created by autonomous LLM agents, or maintained largely through them. The vote was 358 to 144, with 14 abstentions. A lot of IRC software is hosted on Codeberg. Two disclosure-and-provenance decisions in a fortnight is a direction, not a coincidence.

The idea is right; this implementation is not

Disclosure is worth having. People choosing what to run on their own server are entitled to know how it was built, and a support table is a reasonable place to put that.

But a label is only as good as the process behind it, and there isn't one yet. There are no published criteria — "substantially AI-assisted" is a phrase in a pull request comment, not a definition. There is no route for a project to contest a mark, correct it, or volunteer for one. And the most consequential mark so far was applied to a major project without explanation, in a pull request about something else, apparently without asking them.

That is not a scandal. It is a good idea shipped before its governance, by one person acting in good faith with more work in front of them than time to do it — which is how most things in IRC get built. But a flag that carries reputational weight needs criteria published, marks justified where they are not obvious, and a way to be told you got one wrong. Until then, read the robots as "someone believed this" rather than as a finding, and read the empty rows as "nobody has looked" rather than as clean.

See it: the IRCv3 client support tables.