feat: Add OpenAI-compatible API proxy

[chindris-mihai-alexandru]

Summary

Implement an OpenAI-compatible HTTP server mode to allow using Puter models directly in Zed's model dropdown (and other tools that support custom OpenAI endpoints).

Proposed CLI

npx opencode-puter-auth serve --openai --port 3000

Configuration in Zed

"language_models": {
  "openai": {
    "api_url": "http://localhost:3000/v1",
    "available_models": [
      { "name": "puter/claude-opus-4-5", "displayName": "Claude Opus (Puter)" },
      // ...
    ]
  }
}

Implementation Details

• Use express or standard node http.
• Implement /v1/models (list models).
• Implement /v1/chat/completions (streaming and non-streaming).
• Map OpenAI request body to Puter API format.
• Handle authentication via existing auth.ts.

Benefits

• Models appear directly in Zed's dropdown.
• No need to use "tools" or indirect prompting.
• Better UX for users.

[jelveh]

Hi @chindris-mihai-alexandru, thank you for building this great project! We are in the process of finalizing our OpenAI-compatible Puter endpoint, so you won't need proxying. Please stay tuned.

[chindris-mihai-alexandru]

thank you @jelveh for your kind words! looking forward to the update 🚀

[chindris-mihai-alexandru]

Already implemented: OpenAI-compatible proxy is available via Example: puter-auth serve --mcp and documented in README (plus tests). Closing as done.

[chindris-mihai-alexandru]

Correction: OpenAI-compatible proxy runs via puter-auth serve --openai (or npx opencode-puter-auth serve --openai) and is documented in README. The previous comment had a shell substitution typo.