@meego-harness/llm-worker

OpenAI-compatible LLM worker bridge for meego-harness.

This package logs in to a worker server with role llm, receives normal worker text messages, calls an OpenAI-compatible chat completions endpoint, and returns normal worker text responses.

It does not execute manager tools or implement manager-agent-specific metadata protocols. manager-agent owns the worker model prompt, response parsing, and local tool execution loop.

worker-cli usage

The package declares a meegoHarnessWorker manifest and can be installed by @meego-harness/worker-cli as a private package. Its setup action stores local LLM connection settings and returns only worker-cli metadata.

meego-harness-worker onboard llm
meego-harness-worker start --worker llm-worker-1 --tmux

The package bin is also available directly after installation:

meego-llm-worker setup
meego-llm-worker start --worker llm-worker-1

The API key and custom headers are kept in the worker config file and are not included in setup/list/doctor JSON output.

Library usage

import { OpenAICompatibleLlmWorkerBridge } from '@meego-harness/llm-worker'

const bridge = new OpenAICompatibleLlmWorkerBridge({
  serverUrl: 'ws://127.0.0.1:3000/workers',
  workerId: 'llm-worker',
  email: '[email protected]',
  apiKey: process.env.LLM_API_KEY!,
  baseURL: 'https://llm.example.com/v1',
  model: 'model-id',
})

await bridge.start()