phantom
An autonomous AI co-worker with its own VM, persistent memory, and MCP server capabilities.
Autonomous agent with persistent memory, custom tools, and self-evolution on a dedicated machine.
Phantom is an open-source AI agent that runs on its own virtual machine rather than a disposable chat window. It features persistent memory via Qdrant, Slack and web interfaces, and the ability to spin up its own infrastructure. It suits engineers and teams wanting an agent that can build and host functional artifacts.
What it is
The tool gives an AI its own machine to install software, build analytics platforms, monitor its own infrastructure, and extend its own capabilities with new communication channels like Discord. It supports multiple LLM providers including Anthropic, Z.AI, and Ollama through a simple YAML configuration.
Who it's for
It suits software engineers, sales teams, and technical project leads who want a persistent agent capable of self-hosting dashboards, building data pipelines, and executing code. It does not suit non-technical users who cannot manage a Docker deployment or an external LLM API key.
Key features
Dedicated Computer VM
Runs on its own machine to spin up databases, install software, and host web applications.
Persistent Memory Engine
Retains context across sessions using Qdrant vector storage instead of starting from scratch.
Multi-Provider Model Support
Swaps between Anthropic, Z.AI, OpenRouter, Ollama, and vLLM with two lines of YAML.
Self-Evolution & MCP Server
Builds its own tools, registers them as MCP servers, and integrates new communication channels.
Up and running in 3 steps
Our honest verdict
What's great
- Builds and hosts actual web pages, APIs, and dashboards on a public domain
- Integrates with Slack, Telegram, email, and webhooks out of the box
- Swaps underlying LLM providers easily via configuration files
- Extends its own capabilities by writing and registering custom MCP tools
Where it falls short
- Mounts the host Docker socket, granting root-equivalent access that requires a dedicated VM for security
- Requires ongoing API costs for LLM providers unless running local models via Ollama
- Requires manual configuration of environment variables, API keys, and Slack tokens
Pricing & license
Free under the Apache-2.0 licence, with the only direct cost coming from running your own VM and paying for LLM API usage. A managed free hosting option is also provided on ghostwright.dev/phantom.
Open remote AI-training roles
874 open remote AI-training roles, updated November 15, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Can I use local LLMs with Phantom?
Yes, you can configure Ollama to run GGUF models on your own GPU with zero API cost.
Where does Phantom host the applications it builds?
It runs them on its own dedicated VM and public domain rather than trapping them on localhost.
What are the security implications of running Phantom in Docker?
The setup mounts the Docker socket to let the agent spawn sibling containers, so you must run it on a dedicated machine.
