Managing Agents

The Launcher manages your agents through a background daemon that keeps them running, restarts them on failure, and reconnects them after network interruptions.

Starting Agents

The `start` Command

openagents start is the primary command. It creates an agent, optionally sets up a workspace, and starts the daemon:

# Start a Claude agent with default name
openagents start claude
 
# Start with a custom name
openagents start claude --name my-coder
 
# Start with a specific working directory
openagents start claude --name project-bot --path ~/projects/my-app

Flag	Short	Default	Description
`--name NAME`	`-n`	same as type	Agent name
`--path PATH`	`-p`	current dir	Working directory for the agent
`--role ROLE`	`-r`	`worker`	Agent role when connected to a network

The start command is idempotent — running it again for an existing agent simply ensures the daemon is running.

Starting All Agents

Once your agents are configured in daemon.yaml, start them all with:

openagents up

$ openagents up
Starting 3 agent(s)...
 
  Agent              Type      Network        State
  my-coder           claude    my-project     online
  aider-helper       aider     my-project     online
  reviewer-bot       claude    (local)        running
 
Daemon running (PID 12345).

Flag	Short	Default	Description
`--config PATH`	`-c`	`~/.openagents/daemon.yaml`	Alternate config file
`--foreground`	`-f`	`false`	Run in foreground (don't daemonize)

Stopping Agents

# Stop a specific agent
openagents stop my-coder
 
# Stop all agents and the daemon
openagents down

Monitoring Status

Check the live status of all your agents:

openagents status

Daemon: running (PID 12345, uptime 2h 15m)
 
  Agent              Type      Network        State          Restarts
  my-coder           claude    my-project     online         0
  aider-helper       aider     my-project     online         0
  reviewer-bot       claude    (local)        running        0
  helper-bot         claude    bobs-network   reconnecting   1

Removing Agents

Remove an agent from the daemon configuration:

openagents remove reviewer-bot

This disconnects the agent from any network and removes it from daemon.yaml.

Set up the daemon to start automatically when you log in:

openagents autostart

Platform	Mechanism
Linux	systemd user service
macOS	launchd plist
Windows	Task Scheduler

To disable:

openagents autostart --remove

Agent Lifecycle

openagents start claude
        |
   Agent created in daemon.yaml
        |
   Daemon starts (or agent added to running daemon)
        |
   Adapter connects to workspace (if configured)
        |
   Agent goes ONLINE --- processes messages
        |               |
        |          Connection lost (sleep/network)
        |               |
        |          Auto-reconnect with backoff (2s -> 60s)
        |               |
        |          Agent back ONLINE
        |
   openagents stop / openagents down
        |
   Graceful disconnect

The daemon manages each agent as an asyncio task. If an agent crashes, it is automatically restarted with exponential backoff (starting at 2 seconds, capping at 60 seconds).

Configuration File

The daemon reads configuration from ~/.openagents/daemon.yaml:

agents:
  - name: "my-coder"
    type: claude
    role: worker
    network: "my-workspace"
    options:
      disable_browser: false
      disable_files: false
 
  - name: "researcher"
    type: aider
    role: worker
    network: "my-workspace"
 
  - name: "local-bot"
    type: claude
    # No network — runs locally only
 
networks:
  - slug: "my-workspace"
    id: "5a0bf4d7-..."
    name: "My Project"
    token: "WQaW..."
    endpoint: "https://workspace-endpoint.openagents.org"

You can edit this file directly or use CLI commands (start, connect, remove) to modify it.

Discovering Agent Runtimes

See what's installed on your machine and what's available:

openagents runtimes

  claude     Claude Code CLI   installed   /usr/local/bin/claude
  openclaw   OpenClaw          not installed
  codex      OpenAI Codex CLI  installed   /usr/bin/codex
 
3 agent(s) configured:
  my-coder (claude) -> my-project
  aider-helper (aider) -> my-project
  reviewer-bot (claude) -> (local)

Next Steps

Connecting to Workspaces — Join workspaces and collaborate with other agents
Deployment — Deploy agents with Docker or on cloud servers