Dropped the role description (it belongs in the charter), the role-changing note
(not all roles are governance-determined — the charter lets the GM set some), and
the voting-member figure, which was wrong.
Records the membership poll that formally approved Nathan Schneider as general
manager (closed 2026-10-30, 74 members voting), what the role does, and how roles
are added or changed. Linked from the Operations index and the nav.
The Steward Policy passed in Loomio, so it lands as operations/steward-policy.md
(verbatim from the shared draft, one garbled phrase repaired: 'finances alloas
determined by' -> 'finances allow, as determined by').
Governance and Operations now have index pages, and both are in the nav as
section overviews, so each section has a landing page instead of a bare menu.
Same name, new snapshot and new rates ($0.36/$1.45/$0.04 per 1M). The id is unchanged
on purpose: 3,014 calls in two days used it, so renaming would break members' configs.
The gateway now serves four version-free names (deepseek-flash-latest,
glm-flash-latest, deepseek-flash-tee, glm-tee), added per co-op/support#47 so a
member's config survives a provider retiring a version. The example entry is what
members copy, so it should be the names that keep working; provider-prefixed ids
still work and are linked for anyone who needs a specific version.
A member read Regolo's own catalogue, where the input-limits column reads 30K for
apertus-70b (and for the MXFP4 gpt-oss models we do not serve from Regolo), 60K for
mistral-small-4-119b and 100K for gemma4-31b, and asked whether >30K input is
possible at all on Regolo. It is: probed through the co-op gateway, every Regolo
chat model we serve accepts 262,144 tokens and rejects above it, so the ceilings are
exact rather than approximate. gemma4-31b and mistral-small-4-119b therefore move
from their catalogued 100K/60K to the measured 262K.
The 'Context figures are measured' paragraph now says what was measured, when, and
how — and names the Regolo catalogue as the source of the smaller numbers, so the
next reader comparing the two does not file the difference as a bug.
Rates per Regolo's own catalog ($0.40/$0.60 and $0.20/$0.70 per 1M);
context 100K / 60K. Vision verified through the co-op gateway for both.
Mirrors the gateway config (code/litellm config.yaml, authoritative).
pip show exits non-zero for an absent package; under set -euo pipefail the
command substitution aborted build.sh before the self-heal could run — the
gate then failed the deploy with no message at all. Tolerate the missing
package (that is the case being healed) and report it.
The poller's private venv (.docs-publisher/.venv-docs) diverged from the
canonical one twice, each time silently delaying a deploy. Consolidate:
- poll-watcher.sh builds with the ONE canonical venv
(~/repos/docs/.venv-docs) while keeping its dedicated checkout (uncommitted
edits in a working clone must never block or leak into a deploy).
- New drift check: publishing/ + mkdocs.yml are compared with the canonical
checkout before every build; a mismatch fails the deploy loudly instead of
shipping logic the canonical source cannot reproduce.
- Pre-deploy gate: build.sh --check-only verifies the pinned toolchain and
every configured plugin (self-heals missing pins, fails loudly naming an
unpinned plugin).
- Alerting: 3 consecutive deploy failures -> marker file + ONE Matrix message
to the InferenceBot HQ room (hourly cap); counter resets on next success.
- The watcher re-execs itself when a deploy changed poll-watcher.sh, so the
running loop always matches the deployed one.
- webhook receiver (dormant) repointed at the canonical venv too; unit files
and README document the layout.
requirements.txt pinned only mkdocs-material, so a cold venv lacked mkdocs
and mkdocs-redirects — the exact gap that silently stalled the auto-deployer
today. build.sh now gates every build on the venv: it verifies the pinned
requirements (installing anything missing or off-version — fast local no-op
when current) and fails loudly if mkdocs.yml configures a plugin that
requirements.txt does not pin. --check-only runs the gates without building,
for use as the poller's pre-deploy check.
- greenpt/deepseek-v4.1-flash routes to GreenPT's base deepseek-v4.1-flash
id, not the deepseek-v4.1-flash-1017 snapshot (17 Oct 2026) that GreenPT
lists separately. Confirmed from the gateway registry mapping and
GreenPT's model card / model list, 2026-11-09.
- regolo/qwen-image serves Regolo's single Qwen-Image id, which Regolo's
catalog describes as the (2512) checkpoint — the December 2025 update,
not the newer Qwen Image 2.1 (2026-09-20). Regolo does not list 2.1.
- tinfoil/nomic-embed-text is now member-facing ($0.016/1M per Tinfoil's
catalog, TEE-protected, 768-dim, 8K ctx) — the model the co-op's own docs
search runs on inside the TEE.
- greenpt/green-embeddings added ($0.20/1M, 2560-dim measured, 32K ctx) — the
only genuinely distinct other embedding model in GreenPT's catalog.
- Embedding table now lists all three with pricing; quick-guide bullet updated.
- whisper-large-v3-turbo: $0.01 per request (flat, zero per-token) — the old
$0.05/1M-input row was stale (co-op/support#48)
- voxtral-tts: $0.002 per request (was 'listed $0 (verify on invoice)')
- new: tinfoil/voxtral-small-24b, Tinfoil's token-priced STT model
($0.20/1M in + $0.60/1M out), with the chunk-equivalent note for the
transcription route and a chat-with-input_audio example
- note why dashboard audio rows read '$0.01 · 0 tokens' (Tinfoil returns no
token usage for audio)
New gateway model (code/litellm 54925ca): /v1/embeddings, GreenPT-hosted,
4096-dim (measured through the gateway), $0.25/1M input from GreenPT's own
pricing, 32K input cap (provider catalog). Not a chat model — API-only like
the audio models, hidden from the chat picker. Also states plainly that
Tinfoil's nomic-embed-text is the operator-only docs-RAG embedding, since a
member read its absence from the model list as removal.
- condense the page to what a member needs: where services run (Infomaniak),
what backups mean to them, the .coop domain via domains.coop, mail via
Scaleway, the Matrix space and the accounts they may meet
- drop operator detail (migration narration, DNS record inventory, scheduled-task
internals, admin-only components, internal account/credential identifiers) to
co-op/ops-docs; fix a heading glitch where a line began with '#infrastructure'
- brand.css: light-mode .md-tabs uses the header's pale wash #e8f5e8 (was the
saturated --green-deep) with a warm hairline; dark scheme unchanged
Make clear this is the documentation, not the co-op homepage:
- site_name -> 'Inference Cooperative docs' (drives the <title> and the
header brand on every page)
- homepage h1 -> 'Inference Cooperative docs'
- footer gains a 'Co-op homepage' link to https://inference.coop/, added
in the existing copyright.html override alongside the Gitea source link
- mkdocs.yml: list 'Working with the agent' in the nav (a page not in the
nav is hard to find).
- working-with-the-agent.md: the operator token runbook moved out of the
published site into the git-only co-op/ops-docs, so the relative link
(operator-token-runbook.md) would 404; point it at the runbook's new home
in ops-docs, as infrastructure.md already does. Wording unchanged.
- The co-op now runs more than one isolated agent; the page covers all of
them and names infrastructure-bot.
- Disambiguates @inference-bot (automation account) from @inferencebot
(the steward agent members instruct).
- Names the mechanism behind "admitted members": a per-agent allowlist
plus mention-required rooms.
- Restates the deploy norm to match standing practice (bug and security
fixes deploy immediately and are reported; member-visible or risky
changes are confirmed first) and points a stricter norm at the forum.
- Off switch: who can actually reach it, and that stopping is not
restarting.
A first draft of the member-facing norms listed as a design goal on
co-op/support#17: where the agent speaks and listens, what to ask it for,
what not to ask it in a shared room, what it does without confirmation,
what always needs a human, and how it is stopped. Linked from the README
index and the Matrix section of getting-started. Nothing here changes how
the agent is configured.
brand.css and typography.css are referenced at fixed URLs via extra_css, so a
warm browser cache could keep serving the previous stylesheet after a deploy.
Add on_post_build to publishing/hooks.py: at build time copy each extra_css
file to <name>.<sha256-8>.css in the output and rewrite the <link> in the built
HTML. Any content change now yields a new URL. Sources keep their names; hashed
copies exist only in the build output and stale ones are pruned on rebuild.