Agent Runtimes
Detect, install, test, update, and select local code-agent CLIs from the desktop app.
Agent runtimes are the local CLIs that can carry out coding and workspace tasks for the desktop app. The app can detect installed runtimes, show their versions and paths, test their connection, and help you install or update supported CLIs.
Open Agent Runtime Settings
- Open Settings
- Select Agent Runtimes
- Click Rescan to refresh the detected CLIs
- Review each runtime card for install state, version, path, authentication, models, and diagnostics
Installed runtimes can be selected as the default agent runtime. Missing runtimes show install or copy commands when the app knows how to set them up.
Supported Runtime Families
The runtime registry includes local coding agents such as:
| Runtime | Typical Use |
|---|---|
| Claude Code | Claude-backed coding and workspace agent tasks |
| Codex CLI | OpenAI-backed coding tasks with configurable reasoning levels |
| Gemini CLI | Google Gemini-backed local agent tasks |
| OpenCode | Multi-provider local coding agent workflows |
| Cursor Agent | Cursor command-line agent workflows |
| Qwen Code | Qwen-backed coding workflows |
| Devin for Terminal | Devin terminal agent sessions |
| GitHub Copilot CLI | GitHub Copilot command-line coding workflows |
| Pi | Multi-provider local runtime with model and reasoning controls |
Some experimental runtimes may be hidden behind feature flags until they are ready for everyday use.
Install or Update a Runtime
Each runtime card may show one or more setup options:
- Install runs an allowlisted in-app command when the app can do so safely
- Update runs a supported updater or reinstall command
- Copy command copies a terminal command when the setup method should be reviewed and run manually
When a setup option downloads from the network, the confirmation dialog calls that out before the command starts. If a runtime requires Homebrew, npm, Node.js, or another prerequisite, install that prerequisite first.
Select the Default Runtime
- Find an installed runtime card
- Click Select
- Pick the runtime model when the card offers model choices
- Pick the reasoning level when the runtime supports it
The selected runtime is used for new agent tasks unless a task or slash command chooses a different runtime.
Test a Runtime
Click Test on a runtime card to verify that the CLI can start and that authentication appears usable.
Use the result this way:
| Result | What It Means |
|---|---|
| Connection test passed | The app can start the CLI and the runtime responded |
| Connection test failed | The CLI is missing, blocked, unauthenticated, or returned an error |
| Auth unknown | The runtime does not expose a reliable auth probe, or the probe could not determine state |
If a runtime is installed but unauthenticated, sign in through the runtime's own CLI and run the test again.
Troubleshooting
- If a runtime does not appear, click Rescan after installing it
- If the version is wrong, check which binary path the app detected
- If an install command fails, run the copied command in a terminal to see the full package-manager output
- If a model is unavailable, update the runtime or choose Custom model when supported
- If the runtime cannot access the workspace, check workspace permissions and sandbox settings