all-api

Tiny OpenAI-compatible gateway for local coding agents.

It turns local codex, claude, OpenClaw, and Hermes endpoints into a small OpenAI-style base_url + api_key + model service.

Implemented surface:

• GET /v1/models
• POST /v1/chat/completions

Use

From npm:

npm install -g @waibiwaibig/all-api
all-api setup

From source:

git clone https://github.com/waibiwaibig/all-api.git
cd all-api
npm install
npm link
all-api setup

setup asks for the workspace directory, creates an API key, starts the server, in the background, prints the connection details, and exits:

Base URL:
  http://127.0.0.1:4011/v1

API key:
  sk-allapi-admin-...

Models:
  codex-local
  claude-code

Call it:

curl http://127.0.0.1:4011/v1/chat/completions \
  -H "Authorization: Bearer sk-allapi-admin-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "codex-local",
    "messages": [{"role": "user", "content": "Reply with exactly: ok"}]
  }'

Create another key:

all-api key create --models claude-code

Stop the background server:

all-api stop

Notes

• No runtime dependencies.
• No Docker or database.
• Each request starts a fresh agent process.
• Local Codex/Claude adapters run one request per model at a time by default, and all-api only terminates the process group it created for that request.
• Stores API key hashes, not raw keys.
• Codex runs with --sandbox read-only.
• Claude runs with --permission-mode plan.
• Binds to 127.0.0.1 by default.
• /v1 is the OpenAI API version prefix. Use http://127.0.0.1:4011/v1 for OpenAI-compatible clients; root paths like /chat/completions also work.