{{ message }}
Add Responses API compatibility alongside chat/completions #231
Description
Problem
The proxy currently exposes , , and related compatibility routes, but it does not expose OpenAI Responses API routes. This blocks models that are available only through Responses API, such as , even when the underlying GitHub Copilot account can use them.
Proposed change
- add
- add
- forward requests to the upstream GitHub Copilot endpoint
- include verbose logging and token-count reporting similar to the existing chat-completions handler
Related PR
Implemented here: #230
Validation
- source build passes with ℹ tsdown v0.15.6 powered by rolldown v1.0.0-beta.41
ℹ Using tsdown config: /Users/ThangLT4/Desktop/code/copilot-api/tsdown.config.ts
ℹ entry: src/main.ts
ℹ target: es2022
ℹ tsconfig: tsconfig.json
ℹ Build start
ℹ Cleaning 2 files
ℹ Granting execute permission to dist/main.js
ℹ dist/main.js 51.04 kB │ gzip: 12.83 kB
ℹ dist/main.js.map 112.86 kB │ gzip: 27.10 kB
ℹ 2 files, total: 163.90 kB
✔ Build complete in 532ms - locally verified and through
- verified that is not usable via , so Responses support is required
Activity
johnib commentedon Jun 21, 2026
the approach in #230 is right - Copilot exposes a native responses upstream so it's basically a thin passthrough, you forward the body and stream it back. couple gotchas that bit me: mount both
/responsesand the versioned/v1/responses(codex hits the unversioned one depending on config), and setwire_api = "responses"in config.toml or codex quietly falls back to chat/completions and you never hit the new route.i've been running a fork with exactly this for a while (copilot-api-node20 on npm) if you just want it working today - went into it a bit more over on #232. otherwise #230 looks right.