fix(opencode): add GPT routing alongside OSS discovery - #239
Conversation
bc65755 to
b53813f
Compare
|
This stack has been rebuilt on current main with one clean commit per layer. The incremental OpenCode diff is linked in the description; the rejected proxy and broad OSS cohort are gone. Local focused validation is 480 passing with Ruff clean. A maintainer CI approval/review is now the remaining gate. |
b53813f to
3f1c227
Compare
|
Post-review update: OpenCode GPT entries now carry shared context/output limits and prefer the newest eligible GPT; foundation endpoint parsing is defensive against malformed records; CLI fallback tests are deterministic; and the live e2e fixture now populates OpenAI/OSS models and isolates Pi settings paths. Round-3 reviewers found no production blocker; final validation is 588 passed/29 skipped focused and 1062 passed/36 skipped excluding environment-sensitive installed-agent capture tests. |
82e6f84 to
48a91ba
Compare
|
Rebased this PR stack onto current Hard stack dependencyThis is 3 of 3 and must land after #223:
This is a code dependency, not just a review preference: #239's The helper and its tests now live in this PR with their only consumer ( If #217/#223 are squash-merged, please rebase this branch onto the updated Validation
This PR adds lower-level OpenCode GPT routing, but managed-config OpenCode GPT mapping and admin allowlist intersection are deliberately deferred to #290. |
Expose the GLM and Kimi coding-model cohort through Pi and OpenCode with shared token limits and reasoning metadata. Keep unsupported chat models out of discovery, including Inkling until gateway issue databricks#215 is fixed, and retain the GPT-OSS Responses API routing guard.
Centralize Claude family/version parsing so Pi metadata, adaptive-thinking compatibility, and Claude Code's [1m] selector cannot drift. Cover Sonnet 4.5, Opus 4.6, future major versions, Fable fallback, and prefixed model IDs.
`_pi_gpt_model_entry` declared `reasoning: True` without an off-state, so for
the thinking-off case Pi's Responses builder fell back to
`reasoning: {effort: "none"}` (pi-ai openai-responses.js, the
`thinkingLevelMap?.off !== null` branch). `"none"` is only valid on gpt-5.1+,
so every request to gpt-5, gpt-5-mini, gpt-5-nano and gpt-5-5-pro was rejected:
BAD_REQUEST: Unsupported value: 'none' is not supported with the 'gpt-5'
model. Supported values are: 'minimal', 'low', 'medium', and 'high'.
Setting `thinkingLevelMap: {"off": None}` makes Pi omit `reasoning` entirely,
which the gateway accepts for all 14 codex ids. Verified against
/ai-gateway/codex/v1/responses: effort="none" 400s on gpt-5/-mini/-nano/-5-5-pro
and 200s on gpt-5-1..-5-6; omitting `reasoning` is 200 everywhere.
`{"off": "minimal"}` was rejected as an alternative because gpt-5-5-pro 400s on
it too. Same pattern already used for the Gemini 3.x entries.
The rest of Pi's Responses payload was bisected against the gateway and is
fine: store:false, prompt_cache_key, prompt_cache_retention:"24h",
prompt_cache_options, include:["reasoning.encrypted_content"], developer role,
flat tool schemas, and the session_id / x-client-request-id affinity headers.
Regression was hard to spot because the gateway returns
{"error_code","message"} rather than OpenAI's {"error":...}, so Pi's
error-body.js recovery no-ops and every 400 renders as
"OpenAI API error (400): 400 status code (no body)". Reported upstream as
earendil-works/pi#7748.
Refs databricks#286
Configure the Databricks OpenAI Responses provider alongside the validated GLM/Kimi provider, and fall back to foundation-model serving endpoints when UC model services are unavailable.
48a91ba to
816588f
Compare
1 participant
Summary
Adds the OpenCode routing and discovery follow-up on top of #223.
Changes
databricks-openaiprovider so GPT-5, GPT-5.6, and Codex models use the Databricks Responses API alongside thedatabricks-ossprovider.Review scope
This branch is stacked on #223. The incremental change is one commit across the OpenCode/discovery implementation and focused tests:
636c3b6...3f1c227
The parent PR no longer contains the rejected proxy or broad OSS cohort.
Validation
uv run ruff check .— passeduv run pytest --ignore=tests/test_e2e_user_agent.py -q— 1062 passed, 36 skippedThe installed-agent User-Agent capture module was excluded from the full run because local Claude/Pi installations can return from existing managed state without contacting its capture server; all affected routing and rendering paths are covered by the focused tests above.
Replaces the larger consolidated #238 with a clean linear stack.