Skip to content

Foxwarm documentation

Foxwarm is a self-hosted workspace for AI agents that work on long tasks. You run one Foxwarm server. It holds your agents, their conversations and memory, and the model and tool configuration. Agents read code, run commands, hand work to other sessions, and report back in the browser or in a chat app.

  • An Agent is a long-lived identity with its own memory and rules: for example main, reviewer or support.
  • A Session is one conversation of an Agent. An Agent can have many Sessions, and a Session can start child Sessions for parts of a task.
  • A Node is a machine or environment where tools run. The Foxwarm server itself is the master Node; you can add servers, laptops, containers and browser tabs.
  • A Channel connects a chat app such as Telegram, Matrix, QQ, WeChat or WeCom to a Session.

The agent loop, models and credentials stay on the server. Nodes only run the commands they are given. How Foxwarm works explains this split in more detail.

I want to… Read
Install on Linux, WSL, macOS, Windows or Docker Install Foxwarm
Connect a model provider Set up your first model
Keep project knowledge across conversations Agents, Sessions and memory
Work with very long conversations Long conversations and history search
Split work between Sessions or Agents Delegate work
Run tools on another machine or in a browser Nodes
Talk to an agent from a chat app or a team group Channels and group chats
Run something on a schedule Timers
Find my way around the browser interface Using the WebUI
Limit what an agent can do or read Access control
Let another program or Foxwarm instance call in Connect an external MCP client
Back up, upgrade or move my data Data, upgrades and backups

Foxwarm is open source under the MIT license. The GitHub repository has the source, issue tracker and developer documentation.