Phantom

Normal
GitHub TypeScript Apache-2.0

Description

Phantom is an AI co-worker with its own computer, featuring self-evolving capabilities, persistent memory, and MCP server support, autonomously completing complex tasks like a virtual colleague.

Key Features

  • AI co-worker with its own dedicated virtual machine — installs software, builds dashboards, creates APIs
  • Self-evolving capabilities: can extend itself with new tools and communication channels (built Discord from scratch)
  • Multi-provider support: Anthropic, OpenRouter, Ollama, vLLM, LiteLLM, and custom endpoints
  • Persistent memory with Qdrant vector database for long-term context retention
  • MCP server support for tool registration and cross-agent tool sharing
  • Multi-channel communication via Slack, Telegram, Email, Webhook, and self-built channels

Use Cases

💡 Deploy an autonomous AI assistant that builds and manages its own infrastructure
💡 Create data analytics platforms — Phantom built ClickHouse with 28.7M rows autonomously
💡 Set up monitoring and observability for your own infrastructure using AI agents
💡 Build custom tools and dashboards through natural language conversation on Slack

Tags

agent typescript memory mcp automation

Categories

⚡ Agent Tools

Quick Start

1. Docker (recommended): download docker-compose.yaml and .env from the repo.
2. Edit .env with your ANTHROPIC_API_KEY, Slack tokens, and OWNER_SLACK_USER_ID.
3. Run `docker compose up -d` — Qdrant, Ollama, and the agent will start.
4. Check health at http://localhost:3100/health.
5. Alternatively, get a free managed Phantom at ghostwright.dev/phantom.
Visit GitHub

Related Projects