ArcClaw runs a team of 5 specialized AI agents — Team Leader, PD, Frontend, Backend, QA — that collaborate autonomously to turn product requirements into tested, working software. Powered by the Vercel AI SDK.

✨ Features

• 5 AI agents working in parallel — orchestration, design, frontend, backend, QA
• Extensible provider system — OpenAI, Anthropic, DeepSeek, Ollama, or bring your own
• Model tiering — powerful model for reasoning, fast model for code generation
• CLI + Programmatic API — use as a CLI tool or integrate into your Node.js app
• Real-time dashboard — monitor agents, tasks, and messages via SSE
• REST API — submit requirements, manage tasks, query agent status
• File-based persistence — no database required, all data stored as JSON
• File-locking & watchers — safe concurrent writes with proper-lockfile + chokidar

🚀 Quick Start

Prerequisites

• Node.js >= 18
• An LLM API key (or use Ollama for local models)

Install

npm install arcclaw

Configure

cp .env.example .env

Edit .env — set your LLM provider and API key:

LLM_PROVIDER=openai
OPENAI_API_KEY=sk-your-key
LLM_MODEL=gpt-4o
LLM_MODEL_FAST=gpt-4o-mini

Start

npx arcclaw start

Submit a Requirement

curl -X POST http://localhost:3000/api/requirements \
  -H "Content-Type: application/json" \
  -d '{"title":"User authentication","description":"Login + registration with JWT","priority":"high"}'

The Team Leader automatically decomposes the requirement into tasks and assigns them to the appropriate agents.

🤖 Agent Team

| Agent | Role | Model Tier | |---|---|---| | Team Leader | Orchestration, requirement decomposition, task assignment | Powerful (LLM_MODEL) | | PD | Product specs & PRDs | Fast (LLM_MODEL_FAST) | | Frontend | UI components & pages | Fast | | Backend | APIs, services, data models | Fast | | QA | Test plans & execution | Fast |

🔌 LLM Providers

| Provider | Setup | |---|---| | OpenAI | LLM_PROVIDER=openai · OPENAI_API_KEY=sk-... | | Anthropic | LLM_PROVIDER=anthropic · ANTHROPIC_API_KEY=sk-ant-... | | DeepSeek | LLM_PROVIDER=deepseek · DEEPSEEK_API_KEY=sk-... | | Ollama | LLM_PROVIDER=ollama · No key required (local) |

Add custom providers in code:

import { ArcClaw } from 'arcclaw';

app.registerProvider({
  name: 'google',
  apiKeyEnvVar: 'GOOGLE_API_KEY',
  createModel: async ({ model, apiKey }) => {
    const { createGoogleGenerativeAI } = await import('@ai-sdk/google');
    return createGoogleGenerativeAI({ apiKey })(model);
  },
});

📦 Usage

CLI

npx arcclaw start              # Start the agent runtime + API server
npx arcclaw start --port 4000  # Custom port
npx arcclaw providers          # List registered LLM providers

Programmatic API

import { ArcClaw } from 'arcclaw';

const app = new ArcClaw({
  config: {
    llm: { provider: 'openai', model: 'gpt-4o', modelFast: 'gpt-4o-mini' },
    api: { port: 3000 },
  },
});

// Register custom extensions
app.registerTool(myCustomTool);
app.registerProvider(myCustomProvider);

await app.start();

⚙️ Key Configuration

| Variable | Default | Description | |---|---|---| | LLM_PROVIDER | openai | openai, anthropic, deepseek, ollama | | LLM_MODEL | gpt-4o | Team Leader model | | LLM_MODEL_FAST | same | Worker agent model | | API_PORT | 3000 | REST API port | | DATA_DIR | ./data | Runtime data directory | | PROMPTS_DIR | ./prompts | Custom agent system prompts |

Config priority: Programmatic → config file → env vars → defaults

📚 Documentation

Full docs available in docs/:

| Document | Description | |---|---| | Architecture | System design, core modules, agent lifecycle | | Configuration | All config options, env vars, config file | | LLM Providers | Built-in & custom provider setup | | API Reference | REST endpoints, SSE events | | Extending | Custom agents, tools, providers | | FAQ | Common questions & troubleshooting |

Or open docs/index.html for a bilingual (EN/中文) documentation site.

🛠️ Development

git clone https://github.com/zhangfanbin/ArcClaw.git
cd ArcClaw
pnpm install

pnpm run dev          # Start runtime in watch mode
pnpm run dashboard    # Start dashboard
pnpm run dev:all      # Both simultaneously
pnpm run test         # Run tests
pnpm run build        # Compile TypeScript

📄 License

MIT © ArcClaw