vscode-cp-proxy

A proxy server/adapter layer that relays requests to VS Code APIs.

Listens at localhost:5555 by default. For authentication, open output/vscode-cp-proxy for authentication token, or, set your own token using command setToken provided by this extension by pressing (Ctrl+Shift+P) in VS Code.

Currently supported endpoints:

1. OpenAI
   1. /openai/v1/models
      1. Lists the models supported
   2. /openai/v1/chat/completions
      1. Translates the incoming requests in OpenAI format into VS Code chat API invocations.
      2. Streams responses by default. Non-streaming behavior not tested.
      3. Supports tool calling.
      4. Image support.
2. DeepInfra
   1. /deepinfra/flex/openai/v1/chat/completions
      1. Pass-through proxy to https://api.deepinfra.com/v1/openai/chat/completions.
      2. Injects "service_tier": "flex" and "fail_fast": true into the JSON body before forwarding.
      3. Forwards the incoming Authorization header as is. The target host and path are fixed by the extension. Other incoming headers are never forwarded.
      4. Streams the upstream response back to the client unchanged.
      5. The proxy token gate does not apply to this endpoint. The incoming Authorization header carries the DeepInfra API key, so it cannot also carry the proxy token. When the incoming request has no Authorization header, the request is rejected with HTTP 400.

Use with Emacs

1. Load/require vscode-cp-proxy.el
2. M-x vscode-cp-proxy-set-gptel-backend
   1. If using a custom token, you can store it in the authinfo gpg file and configure as:

        (setq vscode-cp-proxy-token (lambda () (cadr (auth-source-user-and-password "vscode-cp-proxy"))))
      

   2. By default, the extension will generate a new token on every launch which can be obtained from output window by selecting vscode-cp-proxy in VS Code.
   3. Choose from available models

demo - usage with emacs

UsageWithEmacs