An AI co-worker that operates autonomously on its own dedicated virtual machine, offering persistent memory and self-evolving capabilities. Phantom addresses the ephemeral nature of current AI agents by providing a persistent workspace where agents can install software, build infrastructure, and continuously improve their performance on user-specific tasks, benefiting both technical and non-technical users seeking a truly integrated AI assistant.

How It Works

Phantom runs as a Bun process on a dedicated VM, isolating the agent's operations from the user's local machine. It leverages a multi-tiered memory system (Qdrant, Ollama) for persistent recall and a novel "Self-Evolution Engine." This engine employs a 6-step pipeline—Observe, Critique, Generate, Validate (using LLM judges), Apply, and Consolidate—to allow the agent to autonomously rewrite its own configuration, adapt to user workflows, and improve over time. Agents can also dynamically create and register MCP tools at runtime, enhancing their capabilities on the fly.

Quick Start & Requirements

The recommended installation uses Docker:

curl -fsSL https://raw.githubusercontent.com/ghostwright/phantom/main/docker-compose.user.yaml -o docker-compose.yaml
curl -fsSL https://raw.githubusercontent.com/ghostwright/phantom/main/.env.example -o .env

Edit .env to include ANTHROPIC_API_KEY, Slack tokens, and OWNER_SLACK_USER_ID. Then run docker compose up -d. Ollama pulls the nomic-embed-text model. RESEND_API_KEY is needed for email sending. A managed, free service is available at ghostwright.dev/phantom.

Highlighted Details

• Built a complete analytics platform from scratch, installing ClickHouse, loading 28.7 million Hacker News rows, and creating a shareable dashboard and REST API.
• Dynamically extended its communication capabilities by building Discord integration on demand.
• Integrated and monitored its own infrastructure using the lightweight Vigil system monitor.
• Outputs are hosted on a public URL with authentication, enabling shareable dashboards, tools, and APIs, unlike agents confined to localhost.
• Self-evolution pipeline includes a 5-gate validation process with LLM judges to ensure safe and effective configuration changes, with versioning and rollback capabilities.

Maintenance & Community

Phantom is part of the Ghostwright ecosystem, which includes Ghost OS, Shadow, and Specter. Contribution guidelines are available in CONTRIBUTING.md.

Licensing & Compatibility

Licensed under the Apache 2.0 license, permitting commercial use, modification, and distribution within closed-source projects.

Limitations & Caveats

The system relies heavily on Anthropic's Claude models and SDK. The self-evolution and dynamic tool-creation features, while powerful, introduce inherent complexity. A dedicated VM is required, incurring monthly costs ($7-20/month suggested), and integration with less established, low-star open-source projects may present stability risks.