Migrating from OpenClaw

arc includes a migration tool that reads an existing OpenClaw installation and converts agents and cron jobs to arc format.

Why migrate?

OpenClaw is a capable agent router, but it requires a TypeScript runtime, a separate process manager, and a JSON-heavy configuration format. arc replaces the same functionality with plain Python and YAML files that you can read and edit without tooling.

What you gain:

Simpler config: YAML instead of JSON, flat file per agent
Standard cron expressions instead of a custom scheduler API
A single Python package with no build step
arc ask from the CLI without starting a separate server

What gets converted

The importer reads ~/.openclaw/openclaw.json and ~/.openclaw/cron/jobs.json.

OpenClaw	arc
Agent definitions (id, workspace, model, description)	`~/.arc/agents/<name>.yaml`
Discord channel bindings (type: route)	`discord.channel_id` in agent YAML
Cron jobs (cron kind)	`~/.arc/cron/jobs.yaml`
Cron jobs (every kind, everyMs)	Converted to `/N * * *` expression

Model name mapping:

OpenClaw model	arc model
`anthropic/claude-sonnet-4-6`	`claude-sonnet-4-6`
`anthropic/claude-haiku-4-5`	`claude-haiku-4-5`
`anthropic/claude-opus-4-7`	`claude-opus-4-7`
Other `anthropic/X`	`X` (prefix stripped)

Running arc import-openclaw

Dry run first

Always do a dry run before writing files:

arc import-openclaw --dry-run

Output:

Dry run -- reading from /Users/you/.openclaw, would write to /Users/you/.arc

  [dry-run] would import agent: coach
  [dry-run] would import agent: trainer
  [dry-run] would import cron job: weekly-plan
  [dry-run] would import cron job: heartbeat
  Skipped: agent:main (already exists)

The dry run reads and parses all OpenClaw files but does not write anything.

Import

arc import-openclaw

Output:

Importing from /Users/you/.openclaw into /Users/you/.arc

  Imported agent: coach
  Imported agent: trainer
  Imported cron job: weekly-plan
  Imported cron job: heartbeat

Done. Review ~/.arc/agents/ and restart the daemon.

Custom OpenClaw directory

arc import-openclaw --from /opt/openclaw

Summary of flags

Flag	Default	Description
`--from`	`~/.openclaw`	Path to the OpenClaw config directory
`--dry-run`	false	Preview without writing

OpenClaw to arc format mapping

Agent

OpenClaw openclaw.json (agents section):

{
  "agents": {
    "list": [
      {
        "id": "coach",
        "description": "Coach Kai",
        "workspace": "/workspace/fitness-coach",
        "model": {
          "primary": "anthropic/claude-sonnet-4-6"
        }
      }
    ]
  },
  "bindings": [
    {
      "type": "route",
      "agentId": "coach",
      "match": {
        "channel": "discord",
        "peer": {
          "kind": "channel",
          "id": 1234567890123456789
        }
      }
    }
  ]
}

Resulting arc ~/.arc/agents/coach.yaml:

name: coach
description: Coach Kai
workspace: /workspace/fitness-coach
system_prompt_files:
  - AGENTS.md
  - IDENTITY.md
  - SOUL.md
  - USER.md
  - TOOLS.md
model: claude-sonnet-4-6
allowed_models: []
permission_mode: approve-all
discord:
  channel_id: '1234567890123456789'

system_prompt_files is populated by scanning the agent's workspace for the standard identity files that exist.

Cron job

OpenClaw cron/jobs.json:

{
  "jobs": [
    {
      "name": "weekly-plan",
      "description": "Weekly training plan",
      "agentId": "coach",
      "schedule": {
        "kind": "cron",
        "expr": "0 19 * * 0"
      },
      "payload": {
        "message": "Write the weekly training plan."
      },
      "enabled": true
    },
    {
      "name": "heartbeat",
      "agentId": "coach",
      "schedule": {
        "kind": "every",
        "everyMs": 1800000
      },
      "payload": {
        "message": "Read HEARTBEAT.md and follow it."
      },
      "enabled": true
    }
  ]
}

Resulting arc ~/.arc/cron/jobs.yaml:

jobs:
  weekly-plan:
    description: Weekly training plan
    schedule: '0 19 * * 0'
    agent: coach
    prompt: Write the weekly training plan.
    notify: discord
    enabled: true
  heartbeat:
    description: ''
    schedule: '*/30 * * * *'
    agent: coach
    prompt: Read HEARTBEAT.md and follow it.
    notify: discord
    enabled: true

Note: everyMs: 1800000 (30 minutes) becomes */30 * * * *. All imported cron jobs get notify: discord by default.

What does not migrate

OpenClaw feature	Status
Non-Discord bindings (webhooks, etc.)	Not migrated, requires manual setup
Custom schedule kinds (other than `cron` and `every`)	Falls back to `0 * * * *` (hourly)
Agent-level permission config	All agents get `approve-all`
`allowed_models`	Not set; add manually if needed
Cron job model overrides	Not set; add manually if needed
Notify per job (discord vs discord_on_urgent)	All get `discord`; edit manually

Post-migration checklist

After running arc import-openclaw:

Review each agent YAML in ~/.arc/agents/ and verify workspace paths exist
Check that discord.channel_id values are correct (they are converted to strings)
Review ~/.arc/cron/jobs.yaml and adjust notify values where discord_on_urgent is more appropriate than discord
Add allowed_models lists to agents where you want to restrict model access
Set model overrides on cron jobs that should use a cheaper model
Verify system_prompt_files lists the files that actually exist in each workspace
Run arc setup if you have not already done so (creates config and directories)
Start the daemon: arc daemon start
Test each agent: arc ask --agent <name> "Hello"
Check arc status to confirm all agents and cron jobs are loaded

Why migrate?​

What gets converted​

Running arc import-openclaw​

Dry run first​

Import​

Custom OpenClaw directory​

Summary of flags​

OpenClaw to arc format mapping​

Agent​

Cron job​

What does not migrate​

Post-migration checklist​