Use your AgentSoul files with OpenClaw

OpenClaw is the most direct home for an AgentSoul soul, because its workspaces are built from the same four files: soul, identity, user and agents. Copy your forged files into the workspace, matching the file names it already uses, and start a new session. The agent then boots with your personality instead of the default one.

How OpenClaw loads a persona

An OpenClaw workspace is a folder of plain files. The persona lives in four of them, loaded automatically at the start of each session. soul.md is the personality core: tone, values and behavioural boundaries, read first. identity.md is the externally visible card: name and role. user.md describes the person the agent serves. agents.md is the operating manual: workflows and capability limits. The default workspace sits in your home folder under .openclaw/workspace, and OpenClaw writes the names in capitals, such as SOUL.md.

Installing your soul

1. Get the files onto your computer. In the AgentSoul app, open your soul, tap “Get the four files”, then “Send to computer”, and choose AirDrop, Save to Files or Mail.
2. Find your workspace folder. If you use several agents, pick the workspace of the agent you want to reshape.
3. Back up the existing persona files by copying them somewhere safe, or into a backup folder outside the workspace.
4. Copy the four AgentSoul files in, renaming them to match the workspace’s convention — soul.md becomes SOUL.md, and so on.
5. If your old SOUL.md had safety rules or red lines, paste them back into a clearly labelled section of the new one.
6. Start a fresh session and ask the agent to introduce itself. It should describe values and a temperament that match your result.

Do not throw away your safety rules

Many OpenClaw users harden their SOUL.md with rules such as treating external content as untrusted, asking before destructive actions, and never pasting secrets into chat. An AgentSoul soul describes character, not security policy. Keep both: the forged personality on top, then a section of non-negotiable rules underneath. Personality decides how the agent sounds and what it cares about; your rules decide what it may never do. The same goes for any lines the agent has written into its own files during earlier conversations — read the backup before you discard it, because some of those notes may be worth carrying over.

Multi-agent workspaces

OpenClaw can run several agents, each with its own workspace, bound to different channels. This is where identity.md earns its place. It carries the display name, role, platform binding and routing context. Two agents that share the same identity and context create routing chaos — messages land with the wrong one. Forge a separate soul for each role, say a calm, patient one for a family group chat and a precise, sceptical one for research, and give each identity.md a distinct name and purpose.

Gao Dao is useful here because it asks first what kind of agent you are configuring: personal assistant, work and study, professional agent, a child’s study buddy, a family calendar, or not sure yet. Choosing “Professional agent” tunes the conversation toward soul files for tools like OpenClaw and Claude Code.

Checking it worked

• Ask the agent what it values most. The answer should echo the ranked values in soul.md.
• Ask how it would like you to give it feedback. It should reflect the preferences in user.md.
• Give it a task with a risky step and see whether it escalates the way agents.md describes.
• If something sounds wrong, edit the file directly. The files are plain text and yours to change.

Remember that the reading behind the files is an AI-generated hypothesis, not a diagnosis. Treat the first week as a fitting: notice where the agent feels like you and where it does not, and adjust.