cliproxy

cliproxy

Register a CLIProxyAPI gateway as an OpenAI-compatible model provider in Oh-My-Pi.

CLIProxyAPI is a self-hosted proxy that exposes upstream models (Gemini, Codex, Claude, Grok — typically via CLI/OAuth accounts) behind the standard OpenAI API (GET /v1/models, POST /v1/chat/completions, Authorization: Bearer <key>, default port 8317). This plugin turns that gateway into the cliproxy provider so its models are usable as cliproxy/<model-id> anywhere in OMP.

Install

bash plugins/cliproxy/install.sh
#   pwsh -File plugins/cliproxy/install.ps1     # Windows
# non-interactive:
bash plugins/cliproxy/install.sh --url=http://localhost:8317 --api-key=YOURKEY

What it does:

1. Prompts for the gateway URL and API key (or reads --url/--api-key, or CLIPROXY_URL/CLIPROXY_API_KEY).
2. Lists the gateway's models (GET <url>/v1/models) to confirm it works.
3. Writes a cliproxy provider into ~/.omp/agent/models.yml with discovery: openai-models-list, so the model list stays current at runtime.
4. Stores the key in ~/.omp/cliproxy.key (chmod 600), referenced from models.yml via apiKey: "!cat …" — never written inline.
5. Loads the provider extension so /cliproxy re-lists models on demand.

Use

# ~/.omp/agent/config.yml
modelRoles:
  default: cliproxy/gemini-2.5-flash
  smol: cliproxy/gpt-5-mini

/cliproxy re-lists the gateway's models and re-registers the provider live.

Notes

• The base URL is normalised to end in /v1 (so http://host:8317 works).
• Re-running preserves an existing cliproxy provider block — delete it from models.yml to regenerate.