Agents

OpenClaw supports multiple isolated agents, each with their own workspace and model. Define them in openclaw.json under agents.list.

Adding an Agent

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16

{
  "agents": {
    "list": [
      {
        "id": "main",
        "workspace": "~/.openclaw/workspace",
        "default": true
      },
      {
        "id": "researcher",
        "workspace": "~/.openclaw/workspace-researcher",
        "model": "anthropic/claude-haiku-4-5"
      }
    ]
  }
}

Then deploy:

1

make deploy

Each agent gets its own isolated workspace directory on the VPS. Seed files into docker/workspace-templates/ to populate the main workspace on first start (no-clobber — existing files are never overwritten).

Exec Approvals

OpenClaw exec approvals use a two-part system:

  1. Policy — ask=off, security=full set in openclaw.json
  2. Socket daemon — started by entrypoint.sh via openclaw node run before the gateway

Without the daemon, the gateway falls back to broadcasting approval requests to WebSocket clients, which time out after 120s. ask=off in the config has no effect without the daemon running.

If execs are still requiring approval after a fresh deploy:

1
2
3
4
5
6
7
8

# Verify policy is loaded
make exec CMD="openclaw approvals get"
# Expected: Defaults | security=full, ask=off

# Verify socket exists
make exec CMD="ls -la /home/node/.openclaw/exec-approvals.sock"
# If missing, check container logs:
make logs