Skip to content

docs: add an Experimental page for direct completions - #54

Merged
PierreLeGuen merged 1 commit into
mainfrom
docs/direct-completions-experimental
Sep 23, 2026
Merged

PierreLeGuen merged 1 commit into
mainfrom
docs/direct-completions-experimental

Conversation

@PierreLeGuen

Copy link
Copy Markdown
Contributor

Summary

Stacked on #49. That PR removes direct endpoints from onboarding and adds experimental direct verification pages, but nothing links to those pages, and the basic usage docs for direct endpoints are gone. Existing direct users would have no page that tells them how the endpoints behave today.

This PR adds one visible Experimental page:

  • cloud/experimental/direct-completions.mdx: a warning banner with the current verification limitations, the endpoint format and registry, a sample request, the route list, links to the experimental verification pages from docs: refresh Cloud verification guides #49, and steps to move to the Gateway.
  • docs.json: a new "Experimental" group at the end of the Documentation tab that holds only this page.
  • cloud/private-inference.mdx: a short "Direct Completions" section that points to the page. It keeps the old /cloud/private-inference#direct-completions anchor working, which the integration guides linked to before docs: refresh Cloud verification guides #49.
  • cloud/verification/index.mdx: a note that points direct-endpoint users to the page.

Limitations in the banner

Checked against code and live endpoints:

  • A direct endpoint's all_attestations contains only the report of the instance that answered: inference-proxy builds the response with vec![report]. Twelve concurrent TLS connections to glm-5-3-flash.completions.near.ai presented three different certificate keys, so a key pinned from one report fails on another instance. Tracked in Direct endpoints: return the complete serving model-attestation set cloud-api#1087.
  • Response signatures live in each instance's in-memory cache (src/cache.rs in inference-proxy), so a lookup that reaches another instance returns 404. The public registry at completions.near.ai/endpoints lists only domain and models, so clients can't target an instance.

Validation

  • mint validate passes.
  • mint broken-links passes.
  • The sample request returns 200 from glm-5-3-flash.completions.near.ai with model: z-ai/glm-5.3-flash.

The base is #49's branch. GitHub retargets this PR to main when #49 merges.

Give direct completions endpoints one visible Experimental page with a
warning banner that states the current verification limitations:
attestation reports cover only the answering instance, and signature
lookups can return 404 on another instance.

Link it from Private Inference (keeping the #direct-completions anchor)
and from the Verification overview, and list it in a new Experimental
navigation group.
@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Sep 23, 2026 •

Copy link
Copy Markdown

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-09-23T09:12:21.599004Z 3104382 PR opened
ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

@PierreLeGuen
PierreLeGuen changed the base branch from codex/refresh-verification-docs to main September 23, 2026 10:04
@PierreLeGuen
PierreLeGuen merged commit 87f8f2e into main Sep 23, 2026
2 checks passed
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.

1 participant