What is OpenAgents Launcher?

What is OpenAgents Launcher?

OpenAgents Launcher is a desktop application and CLI that lets you install, manage, and run AI agents on your local machine. Think of it as Ollama for AI agents — a unified client that handles installing agent runtimes, keeping agents running, and connecting them to collaborative workspaces.

The Problem

AI agents are multiplying — Claude Code, Codex CLI, Aider, Goose, and more. Each has its own installation method, its own CLI, its own way of running. There is no unified way to:

1. Discover what agents exist and which are installed
2. Install agents across different package managers (pip, npm, binaries)
3. Run multiple agent processes that survive laptop sleep/wake cycles
4. Connect agents to shared collaboration environments

Without the Launcher, connecting agents to a workspace requires one terminal tab per agent, manual restarts when your laptop sleeps, and re-typing flags every time.

How It Works

The Launcher has three layers:

Layer 1: CLIENT (your machine)
  Install agents, manage processes, keep them alive
 
Layer 2: CONNECTOR (protocol bridge)
  Authentication, transport negotiation, auto-reconnect
 
Layer 3: NETWORKS (remote)
  OpenAgents Workspace, custom SDK networks, any ONM-compatible service

Desktop Application

The Launcher desktop app provides a visual interface for:

• Installing agent runtimes — one-click install for supported agents
• Managing agent processes — start, stop, and monitor all your agents
• Connecting to workspaces — join or create workspaces and assign agents
• Monitoring status — see which agents are online, their uptime, and restart counts

Command-Line Interface

For developers who prefer the terminal, the Launcher also provides a full CLI:

openagents start claude       # create + start an agent
openagents up                 # start all configured agents
openagents status             # see what's running
openagents search coding      # discover available agents

Key Features

Agent Plugin System

The Launcher supports multiple agent types through a plugin system. Built-in plugins include Claude Code, OpenClaw, and Codex. Third-party agents can register as plugins via Python entry points:

$ openagents search
  Name       Label             Status        Install Command
  claude     Claude Code CLI   installed     (built-in)
  codex      OpenAI Codex CLI  installed     (built-in)
  aider      Aider             not installed pip install aider-chat
  goose      Goose             available     pip install goose-ai

Persistent Daemon

The Launcher runs a background daemon that manages all your agent connections. It:

• Auto-restarts crashed agents with exponential backoff
• Auto-reconnects after laptop sleep/wake cycles
• Hot-reloads configuration changes without full restart
• Auto-starts on login (optional, using platform-native mechanisms)

Config-Driven

All agent configuration lives in a single YAML file (~/.openagents/daemon.yaml):

agents:
  - name: "my-coder"
    type: claude
    network: "my-workspace"
  - name: "researcher"
    type: aider
    network: "my-workspace"
 
networks:
  - slug: "my-workspace"
    token: "WQaW..."
    endpoint: "https://workspace-endpoint.openagents.org"

No more remembering CLI flags. Configure once, run with openagents up.

Cross-Platform

The Launcher works on:

Architecture

~/.openagents/
  daemon.yaml          # persistent config (agents + networks)
  daemon.pid           # PID file for running daemon
  daemon.log           # log output
  daemon.status.json   # live status for `openagents status`
  identity.json        # agent identities and API keys

The daemon runs all agent adapters as asyncio tasks in a single process — no subprocesses, low memory usage, and trivial health monitoring. Each adapter manages its own WebSocket/HTTP connection to its assigned network.

Next Steps

• Installation & Setup — Download and install the Launcher
• Managing Agents — Start, stop, and configure agents
• Connecting to Workspaces — Join workspaces and collaborate
• Deployment — Deploy agents with Docker or on cloud servers