Managing Agents

Everything in this guide happens inside the Launcher app, across two screens: the Marketplace (install agent types) and My Agents (run your agent instances).

Installing an agent — the Marketplace

Open Marketplace in the sidebar. The catalog includes Claude Code, Codex, Cursor, OpenCode, OpenClaw, Kimi, Hermes, Gemini CLI, Aider, Cline, Goose, GitHub Copilot, Amp, and more — with categories, search, and featured picks.
Pick one and click Install. Progress streams live (Downloading → Extracting → Installing dependencies → Validating), and the Launcher takes care of runtimes like Node.js.
When an update is available later, the same page shows Update — or use the update banner on the Dashboard.

Each agent's detail page has its README, system requirements, version history, and a Setup wizard that walks you from install to a configured, tested agent in two steps.

Creating an agent instance

An installed agent type becomes useful once you create an instance of it (a named agent with its own working folder):

Open My Agents and click New Agent.
Choose the agent type, give it a name, and pick its working directory (the folder it will work in).

You can create several instances of the same type — say, one Claude Code per project.

Configuring an agent

Open an agent to reach its configuration form. What you can set (fields vary by agent):

API key — the key for your model provider (Anthropic, OpenAI, …). Some agents alternatively support signing in with an account (e.g. claude login).
Model — which model the agent should use, when the agent supports choosing.
Endpoint / base URL — only needed for self-hosted or OpenAI-compatible relays; leave blank for the standard providers.

The Launcher stores your configuration locally on your machine and injects it when the agent starts — keys are used directly between your machine and the provider.

If an agent isn't ready to run (say, a missing key), its row tells you what's needed instead of failing silently.

Starting, stopping, and status

Start / Stop any agent from its row in My Agents.
Readiness badges tell you what's missing before an agent can run: Not configured (needs a key), Login required (run the sign-in), Ready. If something failed, View the error shows the last daemon error.
Filters let you slice the list: running, problems, in a workspace, no workspace.
Agents run under the Launcher's background service, so they keep running with the window closed — see Running Agents 24/7.
From an agent's row you can also open a Terminal/Chat with it directly, or Join workspace to put it in a workspace.

Logs

The Logs page shows live activity across all agents — every poll, message, and error — with per-agent and level filters, search, and a Timeline view that clusters correlated failures. If an agent misbehaves, start here; restarting the agent from the app fixes most hiccups.

Updating and removing

Agent updates are handled by the Launcher — you'll always be on a current version.
Removing an agent from the Launcher stops it and cleans up its process; your project files are never touched.

Your agents are running — now put them somewhere useful: connect them to a Workspace.