Skip to content
Source
Sign inSign up
TanStack/schemas

OpenCode and OpenCode Go spec sync 404: upstream deleted the Zen route handler files #393

Send to my agent
Closedtombeckenham opened this issue 0 comments
tombeckenhamcommented

Error

ingest_failed (job spec-sync). Both OpenCode providers have been degraded in /v1/status since today's 05:00 to 05:30 UTC shards.

providererrorwhen (UTC, 2026-10-21)
opencodefetch failed: https://raw.githubusercontent.com/anomalyco/opencode/dev/packages/console/app/src/routes/zen/v1/messages.ts → 404 Not Found05:21:09 (colo ATL)
opencode-gofetch failed: …/routes/zen/go/v1/chat/completions.ts → 404 Not Found05:32:53 (colo EWR)

This is not a single transient. Every handler under routes/zen/ now returns 404: v1/chat/completions.ts, v1/responses.ts, v1/messages.ts, v1/models/[model].ts, go/v1/*. So every daily sync will fail the same way until the adapter changes. (On 10-17 and 10-18 opencode failed once on zen/openai/v1/chat/completions.ts. That was a different cause, already fixed by #347.)

Cause (checked against upstream)

On 2026-10-20, upstream commit 772e2ae "chore(console): remove legacy Console leftovers from the website (#57434)" deleted packages/console/app/src/routes/zen/** and lib/inference-proxy.ts. Those files were thin proxies to a separate "new inference" service, and that service is not in the public repo. openCodeRouteSpec (src/server/providers/opencode-routes.ts) fetches those files and requires export function POST, so it now throws on the first row it checks.

The service itself still works: GET https://opencode.ai/zen/v1/models and …/zen/go/v1/models return 200. Only our verification source is gone.

What upstream publishes now:

  • services/www/src/docs/content/console/api/inference.mdx has an Endpoints table: /inference/openai/v1/chat/completions, /inference/openai/v1/responses, /inference/anthropic/v1/messages, /inference/google/v1beta/models/<model>:generateContent.
  • console/models.mdx and console/go.mdx give a per-model endpoint table. Go is now documented at https://opencode.ai/inference/go/{openai,anthropic,google}/..., and the Go table includes a Gemini row (gemini-4-argon-contributor).
  • models.dev api: opencode → https://opencode.ai/inference/openai/v1, opencode-go → https://opencode.ai/inference/go/openai/v1.

Decision needed

  1. Verification source. Replace the handler-file fetch with the published docs (inference.mdx Endpoints table for Zen; the go.mdx endpoint column for Go), hashed as the method source. The other option is to verify against the live host (for example a 401 rather than 404 from an unauthenticated POST). The docs are a stated source. A live probe is not, and it costs a request per route.
  2. Go endpoint ids. schemaEndpointId in opencode-catalog.ts keeps Go ids as v1/..., with the comment "until the documented base is the one to move". The documented base has now moved to inference/go/.... Moving the ids changes the public /v1/schemas/opencode-go/... paths and emits endpoint changes. Keeping them means verifying v1/<tail> against a docs path it does not literally match.
  3. Go Gemini. It is now documented, so the @ai-sdk/google → null rule for Go can be lifted. That can wait for a follow-up.

The hourly triage run did not change code, because 1 and 2 change where route facts come from and could change public endpoint ids. Stored schemas are unaffected: a failed sync leaves the last versions in place.

  1. tombeckenham closed this

Sign in to join this conversation.