Feature · Built
Model routing & cost controls
An LLM router picks the model per project with a fallback chain. Gemini Flash-Lite is the default. A daily token budget per project switches to email capture when it runs out.
Cheap models first, fallbacks, and a daily budget.
Router with fallbacks#
Each project sets a primary model, a fallback and an embedding model. The defaults are gemini-3.5-flash-lite, then gemini-3.1-flash-lite, and gemini-embedding-001 for embeddings. An optional escalation model takes turns where the facts only match weakly. Providers sit behind an adapter, and a fake provider makes tests and offline development free.
Daily token budget#
Set daily_token_budget on a project. When it's spent, the agent stops calling the model and offers email capture instead, so cost can never run away.