OpenCues is model agnostic. Supported providers today: Cerebras (the recommended default), Groq, OpenAI, Anthropic, Gemini, OpenRouter, OpenCode Zen, and local models via Ollama. You bring your own API key, or connect through an existing Claude or OpenAI subscription; there is no OpenCues account or middleman service.
Why Cerebras is the default
Inline assistance lives or dies on latency: a suggestion that arrives after you have moved on is noise. Cerebras serves the same open-weight gpt-oss-120b family as Groq with lower time-to-first-token on the free tier, and OpenCues' prompts are designed so its prefix cache keeps repeat calls fast.
Per-feature routing
You can route different features to different providers with per-feature settings buckets: one for cues, one for blanks. A common setup is a fast cheap model for word cues and a stronger model for rewrites. Providers that train on your input are refused for prose-bearing features by design.
Local models
The Ollama provider runs everything against models on your own machine: no text leaves your computer, at the cost of whatever latency your hardware gives. Small fast models suit cues; blanks and rewrites want the strongest model you can run.
Setup is an environment variable per provider (for example CEREBRAS_API_KEY); see the README for names and switching. Related: the security model · Supported models · Rate limits · GitHub repo.