feat(plugin): rewrite codex request body for ChatGPT backend requirements - #39197
Open
patrickpassosb wants to merge 2 commits into
Open
feat(plugin): rewrite codex request body for ChatGPT backend requirements#39197patrickpassosb wants to merge 2 commits into
patrickpassosb wants to merge 2 commits into
Conversation
…xApiEndpoint Read provider-level codexApiEndpoint (and a baseURL ending in /codex, with /responses appended) from the auth loader's provider info so a ChatGPT Plus/Pro/Team subscription can be routed through a gateway instead of the hardcoded chatgpt.com backend endpoint. The rewrite is byte-identical to today when neither option is set: the CODEX_API_ENDPOINT constant is used. Refactors the endpoint resolution into an exported resolveCodexApiEndpoint helper and threads the second loader argument (provider info) into the existing URL rewrite site. The experimental websocketFetch path receives the rewritten URL via the same variable, so both transport modes honor the config.
…ents The ChatGPT codex backend (chatgpt.com/backend-api/codex/responses) rejects requests unless store=false, stream=true, and max_output_tokens is omitted. Add a request-body rewrite in the CodexAuthPlugin fetch wrapper that is applied only when the request is routed to the codex endpoint; non-codex OpenAI requests are left byte-identical.
Contributor
|
The following comment was made by an LLM, it may be inaccurate: Related PR found:
No duplicate PRs found. PR #39197 is a unique feature that extends functionality from #38903. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Issue for this PR
Follow-up to #38903. #38903 added the
codexApiEndpointoption so ChatGPT Plus/Pro (OAuth) inference can route through a gateway instead of the hardcodedhttps://chatgpt.com/backend-api/codex/responses. That PR routes the URL correctly, but opencode's standard request body violates the ChatGPT codex backend's hard requirements and every request is rejected. This PR rewrites the request body for the codex path so opencode can actually drive a ChatGPT subscription through a gateway end-to-end.Type of change
What does this PR do?
The ChatGPT codex backend (
chatgpt.com/backend-api/codex/responses) has hard requirements that the OpenAI platform API (api.openai.com) does not:"store": falseis required (else{"detail":"Store must be set to false"})."stream": trueis required (else{"detail":"Stream must be set to true"})."max_output_tokens"is unsupported and must be omitted (else{"detail":"Unsupported parameter: max_output_tokens"}).Verified live (2026-07-26) with a hand-crafted request through a Tailscale Aperture passthrough provider using a ChatGPT Plus OAuth token: a request succeeds (streams a real
response.createdSSE event) only when all three hold. Without this PR,opencode run --model openai/gpt-5.6-solhangs/fails even though the URL routing from #38903 works.This PR extends the
CodexAuthPluginfetch wrapper (the same one that rewrites the URL viacodexApiEndpointfrom #38903) so that when the request is headed to the codex endpoint, the JSON request body is rewritten to:store: false.stream: true(the AI SDK already handles streaming responses, so the response is passed through unchanged).max_output_tokens.The rewrite is guarded: it only applies when the URL is the codex endpoint (the
codexApiEndpointpath), NOT for the standardapi.openai.com//v1/responsespath. Non-codex OpenAI requests are byte-identical.How did you verify your code works?
bun typecheck(tsgo --noEmit) frompackages/opencode— clean.bun test test/plugin/codex.test.ts— 35 pass / 0 fail, including 7 new body-rewrite tests covering: (a)store:falseset on codex path, (b)stream:trueset on codex path, (c)max_output_tokensstripped on codex path, (d) non-codex requests unchanged, (e) body rewrite only applies when URL is the codex endpoint.opencode run --model openai/gpt-5.6-sol "Reply with exactly: CHATGPT PLUS OK"returned a real streamed answer through a Tailscale Aperture passthrough provider. Non-codex model (ollama-cloud/glm-5.2) still works unchanged.Screenshots / recordings
N/A — no UI change.
Checklist