What are Workspaces?
Workspaces are the isolated environments where Claws operate. Learn about workspace files, configuration, memory, skills, and how workspaces keep agents separate.
What is a Workspace?
A Workspace is the isolated environment where a Claw operates. Every Claw gets its own dedicated workspace containing configuration files, memory, installed skills, uploaded documents, and session history. Workspaces ensure that each agent has its own context and cannot interfere with other agents on the team.
What files make up a Workspace?
Each workspace is structured as a directory with several key files that control how the Claw behaves:
| File | Purpose |
|---|---|
AGENTS.md | Operating instructions -- rules, priorities, and high-level guidance for how the Claw should behave. |
SOUL.md | Personality and tone -- the Claw's communication style, boundaries, and identity foundation. |
MEMORY.md | Long-term memory -- curated knowledge the Claw retains across sessions. |
USER.md | User context -- information about the user the Claw works with, including preferred name and relationship details. |
These files are loaded into every session, giving the Claw consistent context regardless of when or how it is engaged.
How do operating instructions work?
The AGENTS.md file is the primary way you tell a Claw what to do and how to do it. Operating instructions can include:
- Priorities -- what matters most when the Claw makes decisions.
- Rules -- constraints on behavior (e.g., "always get approval before sending emails").
- Domain context -- background information about your business, industry, or workflow.
- Memory usage rules -- how the Claw should store and reference information.
You edit these instructions through the Operating Instructions settings page, and changes are written to AGENTS.md automatically.
How does personality configuration work?
The SOUL.md file defines the Claw's communication style. This includes tone (professional, friendly, concise), boundaries (what topics to avoid), and identity (how the Claw introduces itself). See Personality for more on configuring this file.
How does workspace memory work?
The MEMORY.md file stores long-term knowledge that the Claw builds up over time. Unlike chat history which is session-based, memory persists across conversations. The Claw can add to its memory automatically based on interactions, and you can also edit memory directly.
Workspace memory is separate from the session-level context window. It acts as a curated knowledge base that the Claw can reference in any conversation. See Agent Memory for details.
What about files and documents?
Workspaces support uploaded files and a shared drive. You can upload documents, spreadsheets, images, and other files for a Claw to reference during conversations. These files live within the workspace and are only accessible to that specific Claw.
For files that need to be shared across the team, use the Shared Drive feature.
How are skills stored in a Workspace?
When you install a Skill on a Claw, the skill's configuration is registered within the Claw's workspace. This means different Claws can have different skill sets, even on the same team. The workspace tracks which skills are active, their configuration, and any requirements they depend on.