Skip to main content

Your Agent's Environment

Each agent runs in its own Docker container — a real Linux box with a filesystem, a shell, and common dev tools. The container can be recreated, but two directories persist across restarts:

  • /workspace — the agent's private storage. Memory file, project files, shell init, and local working state live here.
  • /shared — mounted across every agent in the same tree. This is how agents share repos, data, and knowledge bases.

Everything else resets when the container is recreated. If you need something to survive, put it in /workspace or set it up in /workspace/.shell-init.sh, which runs automatically at the start of every shell session.

Editor and Terminal​

When you select an agent, two tabs give you direct access to its environment:

  • Editor — browse and edit files in the agent's container.
  • Terminal — a live shell session into the same container the agent uses.

These are useful when you want to verify what the agent actually did, fix a file yourself, or inspect the state before giving the next task.

MEMORY.md​

The file at /workspace/MEMORY.md is special. Taurus loads it into the agent's context at the start of each run, so this is where long-lived reminders and working context belong.

You can edit it directly in the Editor tab. If an agent seems confused or is acting on stale assumptions, fixing MEMORY.md is often the fastest intervention.

What's next​

Learn how agents coordinate with each other in Multi-Agent Collaboration.