proxyllm-bridge

proxyllm-bridge

OpenAI-compatible bridge server for routing ProxyLLM requests to a local Claude Code CLI session.

The bridge exposes:

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

At runtime it uses only Node built-ins and calls local claude -p with tools disabled by default:

claude -p "<prompt>" --tools "" --safe-mode --no-session-persistence --output-format json

Install

npm install -g proxyllm-bridge

Run with auth:

export PROXYLLM_BRIDGE_TOKEN="$(openssl rand -hex 32)"
proxyllm-bridge serve --host 0.0.0.0 --port 8787

For development only, auth is disabled when PROXYLLM_BRIDGE_TOKEN is unset and NODE_ENV is not production.

NODE_ENV=development proxyllm-bridge serve

On startup the bridge prints its public Bridge URL (http://<public-ip>:8787). Copy that URL into the ProxyLLM dashboard to connect the bridge to your account.

systemd setup

On Linux, this writes a user systemd service and an env file:

proxyllm-bridge setup

For a system service:

sudo proxyllm-bridge setup --system

After setup, run the commands printed by the CLI. The env file includes:

PROXYLLM_BRIDGE_TOKEN=<generated shared secret>
PROXYLLM_BRIDGE_HOST=0.0.0.0
PROXYLLM_BRIDGE_PORT=8787
NODE_ENV=production

Examples

Health:

curl http://127.0.0.1:8787/health

Models:

curl -H "Authorization: Bearer $PROXYLLM_BRIDGE_TOKEN" \
  http://127.0.0.1:8787/v1/models

Chat completion:

curl -s http://127.0.0.1:8787/v1/chat/completions \
  -H "Authorization: Bearer $PROXYLLM_BRIDGE_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-code",
    "messages": [
      { "role": "system", "content": "Answer concisely." },
      { "role": "user", "content": "Say hello from ProxyLLM." }
    ]
  }'

Configuration

• PROXYLLM_BRIDGE_TOKEN: bearer token required when NODE_ENV=production
• PROXYLLM_BRIDGE_HOST: bind host, default 127.0.0.1
• PROXYLLM_BRIDGE_PORT: bind port, default 8787
• PROXYLLM_BRIDGE_MODEL: model id returned by /v1/models, default claude-code
• PROXYLLM_BRIDGE_CLAUDE_TIMEOUT_MS: Claude CLI timeout, default 120000
• PROXYLLM_BRIDGE_MAX_BODY_BYTES: max JSON request body, default 1048576
• CLAUDE_BIN: Claude CLI executable, default claude

Notes

The bridge converts OpenAI chat messages into a plain text transcript before passing it to Claude Code. stream: true returns a compatible server-sent event response after Claude finishes, but it does not stream tokens live from Claude.