Skip to content

docs: reference pages match the gateway users actually call (issue #68) - #74

Merged
SebConejo merged 3 commits into
mainfrom
fix/issue-68-reference
Aug 18, 2026
Merged

SebConejo merged 3 commits into
mainfrom
fix/issue-68-reference

Conversation

@SebConejo

Copy link
Copy Markdown
Member

Applies the recommended options for all ten findings of #68 (batch pass, 2026-08-18), deduplicated against the already-merged PRs #62 and #65:

  1. Rate limits: the section states the real fixed caps (200/min per workspace, 500/min per IP, 10 in flight — same on Cloud and self-hosted, no setting moves them), and the THROTTLE_* rows are rescoped to the dashboard's own API calls. When platform PR #2724 (env-configurable caps, still open) is merged and released, a follow-up documents the real overrides.
    2-3. 424 and the error envelope: the api.mdx and headers.mdx halves were already fixed by docs: llm-gateway and API reference drop the 424 myth (issue #53) #62; this PR fixes the glossary's last 424 and adds the provider-forwarded-429 clause to the status table.
  2. Tier and reason headers: live-verified on a local instance — a Manifest-answered request reports X-Manifest-Tier: simple with model and provider manifest, values the old rows said could not occur. Rows rewritten to the values a client can receive; the tier's name appears in the dashboard, never in a header.
  3. Body forwarding: "forwarded verbatim" becomes what Manifest does — converts per provider, renames max_tokens, merges harness defaults, drops OpenAI-only fields outside OpenAI/OpenRouter routes.
  4. Already fixed by docs: LAN exposure goes through HOST_BIND_ADDRESS in .env (issue #56) #65 (nothing to do).
  5. anthropic-version: accepted and ignored; Manifest sets its own upstream version.
  6. Header beats model ID: a matching custom-tier header takes precedence over an explicit model ID — stated on the glossary, headers page, and (parked item, same wording) the routing page.
  7. Client-kind split: chat/streaming clients get errors as 200-with-message, stated above the status table; the headers section heading stops promising "every request".
  8. M302 softened to the connected-provider condition, matching the other two pages.

Also consumes the parked self-hosted.mdx line (drops "rate limits" from the optional-env list).

Coherence: GREEN. Closes #68.

The rate-limit section states the real fixed caps (200/min per
workspace, 500/min per IP, 10 in flight, identical on Cloud and
self-hosted) and the throttle env rows are rescoped to the dashboard's
own API calls. The glossary's last 424 goes. The tier and reason header
rows document the values a client can receive — live-verified: a
Manifest-answered request reports tier simple with model and provider
manifest. The verbatim-forwarding claim becomes what Manifest does to
the body. The Anthropic version header is accepted and ignored. A
matching custom-tier header beats an explicit model ID, stated on the
glossary, headers, and routing pages alike. Chat clients' 200-with-
message delivery heads the status table, and the headers section stops
promising every request.
@SebConejo
SebConejo merged commit 5524478 into main Aug 18, 2026
1 check passed
@SebConejo
SebConejo deleted the fix/issue-68-reference branch August 18, 2026 22:08
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Docs audit — reference (api, headers, environment variables, glossary) — 2026-08-16

1 participant