Skip to main content
An Environment is everything an agent needs to run, packaged once and connected to as many agents and assistants as you like. Without it, each agent carries its own copy of the same setup: the same MCPs, the same documents, the same API keys, the same terminal configuration. Change something and you change it in ten places, or in nine and forget the tenth. Environments

​
What an environment carries

Tools

The MCPs and tool groups every connected entity gets.

Knowledge

Documents and folders the connected agents can read.

Environment variables

Values available at runtime, encrypted at rest and shown masked.

Sandbox

A real Linux machine: runtime, init script, installed CLIs and its snapshot.

​
How connecting works

An agent or assistant connects to one environment; an environment serves many. The connection is additive: what the environment brings is added to what the entity already had, so connecting one never takes tools or documents away. The connection itself can carry variables of its own, for values that differ per agent while everything else is shared. Precedence runs from most specific to least: connection → environment → sandbox.
The sandbox is the exception to “additive”. When an environment with a sandbox is connected, the environment governs the sandbox completely — the entity’s own terminal configuration is ignored at runtime, not merged. The interface makes it read-only and says so, because a setting that looks editable but has no effect is worse than one that is visibly locked.

​
When to use one

​
Next steps

Configuration

Variables, secrets, tools and knowledge.

Sandbox

The machine agents run commands on, and its snapshots.

CLIs

Command-line tools baked into the snapshot.

Projects

Environments belong to a project, like everything else.