Quickstart
Build agentdesktop and choose standalone local YAML or a controller-managed deployment for AI developer tool configuration.
Start with Build and install when working from source. The quickstarts then cover two configuration modes:
- Standalone reads desired configuration from local YAML. It creates no controller-managed device identity and can use OIDC to authenticate the user directly to an inference gateway.
- Controller-managed enrolls the user and device, receives versioned configuration from the controller, and can request short-lived controller-signed gateway JWTs.
Both modes reconcile the supported developer tool’s own configuration. The inference gateway is a separate service; the source repository includes local and Kubernetes agentgateway examples.
Build and install
Build and install the agentdesktop desktop app, CLI, device daemon, controller, and embedded web interfaces from source.
Standalone
Run agentdesktop from local YAML, discover AI developer tools, and connect directly to an inference gateway without a fleet controller.
Controller-managed
Enroll agentdesktop devices with a controller, distribute managed AI tool configuration, and issue short-lived inference gateway credentials.