Skip to main content

Connect an AI provider

Lotti separates an AI setup into three layers: a provider is the service it connects to, a model is one model exposed through that service, and an inference profile routes each kind of work to a model. Start with the provider, then review the models and profiles created around it.

All screenshots
AI Settings keeps cloud and local connections together. Each card shows connection state and how many saved models use it.
ScreenshotMobile view
Lotti AI Settings showing Mission Control Router, Habitat Local Lab, Orbital Vision, and Penguin Audio Bay providers

Choose local or remote​

  • A cloud provider is usually the quickest setup and commonly requires an API key. Text, audio, or images sent through it leave the device.
  • A local provider can keep inference on your own machine or a local endpoint you control, but its service must be running and reachable from the device where Lotti invokes it. A LAN address may still move a prompt between your devices; “local” means it is not sent to a hosted model provider, not that no network hop exists.
  • A custom base URL can point a supported provider type at a compatible proxy or gateway. The provider type still controls protocol-specific behavior.

Connect a provider​

  1. Open Settings → AI Settings → Providers.
  2. Select the + button.
  3. Pick a supported provider.
  4. Enter its display name, endpoint, and API key when required.
  5. Add any suggested models you want, then save.

For supported first-run providers, Lotti seeds useful models and one or more inference profiles. The result screen summarizes what was created.

All screenshots
Credentials stay masked. Suggested models are shortcuts, not a requirement—you can add provider-specific model IDs later.
ScreenshotMobile view
Lotti provider editor for Mission Control Router with OpenRouter selected, a base URL, masked API key, and suggested models

Inspect and maintain a connection​

Open a provider card to inspect its masked credential, base URL, saved models, and any active profile that depends on those models. From here you can edit the connection, add another model, or remove the provider.

All screenshots
The dependency summary makes the blast radius visible before you change or remove a provider.
ScreenshotMobile view
Mission Control Router provider detail showing its connection, Waddle Command models, and active Project Waddle profile

Changing a base URL or API key affects every model attached to that provider. After changing either value, verify a representative AI action before assuming the connection is healthy. Removing a provider can also affect profiles that route work through its models; review the dependency information first.

Review profile routing​

Open Models and inference profiles to verify what the provider exposes and how Lotti routes each capability. Categories can set a default profile for their work, while individual automation can select or override a model where the workflow offers that choice.

Privacy boundary​

Before sending sensitive text, audio, or images, confirm whether the configured endpoint is local or remote. A friendly name such as “Habitat Local Lab” is not the security boundary—the base URL, the network path, and the machine serving it are. Lotti stores the logbook locally either way; this decision controls the inference call itself.

API keys

Treat provider keys as secrets. Never paste them into documentation, issues, or screenshots.