One Agent

A modular AI agent framework with multi-LLM support, tool execution, browser/desktop control, messaging platform integration, and self-evolving skills.

Features

• Multi-LLM: OpenAI, Anthropic, Google AI, and any OpenAI-compatible API (Ollama, OpenRouter, LiteLLM)
• Tool Execution: Shell commands, file operations, browser automation (Playwright), desktop control (macOS)
• Messaging: Feishu (Lark) and QQ bot integration via WebSocket
• Memory: Three-layer system — raw conversations, LLM-condensed summaries, knowledge graph
• Skills: YAML-defined reusable workflows with auto-creation, auto-refinement, and auto-pruning
• Multi-interface: CLI, REST API (Fastify), WebSocket
• Multi-process: Each agent runs in an isolated child process

Quick Start

npm install one-agent

CLI Mode

# Set your LLM API key
export OPENAI_API_KEY=sk-...
# Or use Ollama (no key needed)

npx one-agent

Web API Mode

npx one-agent --web
# Server starts on http://0.0.0.0:3000

# Create an agent
curl -X POST http://localhost:3000/api/agents \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-4o"}'

# Chat
curl -X POST http://localhost:3000/api/agents/{id}/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "List files in current directory"}'

Using with Ollama

Edit config/default.yaml:

llm:
  defaultModel: "proxy/gemma4:latest"
  providers:
    proxy:
      baseUrl: "http://localhost:11434/v1"
      apiKey: "ollama"

Configuration

See config/default.yaml for all options including LLM providers, plugin settings, auth defaults, and web server config.

Architecture

Agent Manager → Agent Process (independent child process)
  ├── LLM Router (OpenAI / Anthropic / Proxy)
  ├── Memory Manager (raw → condensed → knowledge graph)
  ├── Tool Registry
  │   ├── Shell Plugin (exec, read, write, list)
  │   ├── Browser Plugin (navigate, click, type, screenshot, extract)
  │   ├── Desktop Plugin (screenshot, mouse, keyboard)
  │   ├── Feishu Plugin (send, reply, cards)
  │   └── QQ Plugin (private, group, reply)
  └── Skill Engine + Evolver

License

MIT