Artificial

Artificial

An open source AI agent harness for orchestrating multiple AI workers from a single dashboard. Built in Go.

Artificial lets you spawn, manage, and coordinate AI agents (Claude Code, OpenAI Codex, ACP-compatible models, local LLMs) as a team. Agents get personas, skills, communication channels, and a shared task board. You manage them from a real-time web dashboard or the REST API.

What it does

• Multi-agent orchestration — run multiple AI agents in parallel, each with their own role and persona
• Real-time dashboard — chat with agents, manage tasks on a kanban board, monitor activity
• Channel-based communication — agents can message each other and receive notifications mid-task
• Multiple backends — supports Claude Code, OpenAI Codex, ACP (Agent Communication Protocol), and local models via OpenAI-compatible APIs
• MCP integration — each worker exposes tools to its agent via Model Context Protocol
• Session persistence — stop and resume agent sessions without losing context

Architecture

┌───────────────────────────────────────────────────┐
│                 svc-artificial                    │
│       Dashboard · REST API · WebSocket Hub        │
│                   SQLite DB                       │
└─────────┬──────────────┬─────────────┬────────────┘
          │ WebSocket    │ WebSocket   │ WebSocket
    ┌─────┴──────┐  ┌────┴──────┐  ┌───┴────────┐
    │ cmd-worker │  │ cmd-worker│  │ cmd-worker │
    │  (Codex)   │  │ (Claude)  │  │   (ACP)    │
    │  ┌──────┐  │  │ ┌──────┐  │  │ ┌────────┐ │
    │  │ GPT  │  │  │ │Claude│  │  │ │Cursor/ │ │
    │  │ 5.4  │  │  │ │Opus  │  │  │ │opencode│ │
    │  └──────┘  │  │ └──────┘  │  │ └────────┘ │
    └────────────┘  └───────────┘  └────────────┘

Quick start

# Build both binaries
make build

# Start the central service (dashboard at http://localhost:4000)
make run-artificial

# In another terminal, start a worker (or spawn from the Dashboard)
make r