OPENProviders and models
Connect a local engine, an API key, or a ChatGPT sign-in — and see only the models you actually have.
The provider and model registry is generated from a snapshot of the models.dev catalogue — roughly 150 providers and several thousand models — so the picker lists what exists rather than a hand-kept shortlist. Only providers you have connected show their models.
Ways to connect
| Method | What it means |
|---|---|
| Local | Ollama runs models on your machine with no key, no account, and no network call |
| API key | Connect a provider in Settings → Providers. Only connected providers' models appear in pickers |
| ChatGPT sign-in | OpenAI accepts either an API key or a ChatGPT sign-in (PKCE OAuth), for the model set that backend serves |
Generation providers
| Job | Providers with a real adapter |
|---|---|
| Speech | ElevenLabs, OpenAI |
| Video | Google Veo, OpenAI Sora |
Providers without a real adapter stay listed but honestly unavailable rather than pretending to work. Google Veo supports image-to-video from a reference image; Sora reference images need a multipart upload path this build does not send.
Choosing which models appear
Settings → Models searches the catalogue and controls which models show up in pickers, so a connected provider with hundreds of models does not flood the list. Settings → Edit Agent sets the model the agent uses by default.
Working with no provider at all
Editing, the timeline, and export need no provider. A local Ollama model is enough for the agent, though watching footage needs a vision-capable model. The catalogue is regenerated in the repository with scripts/generateLlmCatalog.mjs.