nixpkgs: self-manage by 2026-06-09 (current + future versions auto-online) or cancel the channel

[Arthur-Ficial]

Goal

Get the current version online and every future version online, automatically, managed by us - on the nixpkgs channel. One decision, not a pile of tickets. Either nixpkgs becomes a self-service channel we control, or we drop it.

The deadline: 2026-06-09 (14 days)

By then, all four of these must be true, or we cancel the nixpkgs distribution path:

1. Current version live - NixOS/nixpkgs#522554 merged, so nixpkgs#apfel-llm is at 1.3.8 (matching Homebrew + the latest release).
2. We are the maintainer - NixOS/nixpkgs#524394 merged (adds arthurficial as the apfel-llm maintainer).
3. On the team - arthurficial is a member of @NixOS/nixpkgs-maintainers.
4. Self-merge proven - we have merged at least one apfel-llm bump ourselves via @NixOS/nixpkgs-merge-bot merge (no waiting on a random committer).

If all four hold, nixpkgs is self-managed: future releases auto-open a single bump PR (scripts/publish-nixpkgs-bump.sh, already fixed to advance one PR) and we merge it ourselves in minutes. Done. No more nixpkgs tickets.

Why this is the only nixpkgs lever we don't already control

The bump automation works - it opens a correct PR every release. The lag is purely nixpkgs merge latency: only a committer can merge, and the two prerequisite PRs above each need one committer merge to bootstrap. That bootstrap is the single thing outside our control. Hence the hard deadline instead of waiting indefinitely.

Decision on 2026-06-09

• All four true -> keep nixpkgs, close this ticket. Channel is self-service from here on.
• Not all true -> cancel the nixpkgs channel:
  • drop the nixpkgs bump step from make release (remove the publish-nixpkgs-bump.sh call),
  • remove nix profile install nixpkgs#apfel-llm from README + docs/ install methods,
  • leave the package to r-ryantm / community (or open a removal PR if we want it gone),
  • supported channels become Homebrew core + Arthur-Ficial tap + direct download - the ones we fully control.

Either way the user-facing promise ("the free AI already on your Mac, installable") is unaffected - Homebrew + tap already deliver current and future versions today.

cc @franzenzenhofer

[Arthur-Ficial]

Clean framing. One decision, four acceptance criteria, a hard deadline - this is the right way to resolve the nixpkgs merge-latency drag.

The automation side (scripts/publish-nixpkgs-bump.sh) already works and the single-PR-per-version fix landed. The only open variable is the two upstream PRs (NixOS/nixpkgs#522554 for the version bump, NixOS/nixpkgs#524394 for the maintainer addition) and the resulting team membership. All outside our repo, all requiring a nixpkgs committer.

Labelling as enhancement. Nothing actionable from our side until the upstream PRs move - Franz decides on June 9th.

Cheers, Arthur

cc @franzenzenhofer

---

Generated by Claude Code

[Arthur-Ficial]

Status update (2026-06-01): the bump automation is firing correctly - v1.4.0 and v1.5.0 each advanced the single bump PR NixOS/nixpkgs#522554, now at 1.5.0. The four bootstrap conditions (PR #522554 merged, maintainer PR #524394 merged, @NixOS/nixpkgs-maintainers membership, one self-merge) all require an upstream nixpkgs committer to merge - that's the one lever outside our control, so I can't close those out autonomously. This is your call on the 2026-06-09 deadline: ping a committer / @NixOS/nixpkgs-merge-bot to bootstrap, or decide to cancel the channel. Everything on our side is ready.