All posts
Engineering
7 min read
Routing every question to the right model
No single model is best at everything. Orrery's router picks the right one for each question, quietly and in milliseconds.
Daniel Kovacs

Some questions need careful reasoning across months of context. Others need a fast, cheap answer. Treating them the same wastes time, money or both.
How the router decides
Each request is scored on the context it needs, the precision it demands and how quickly you need an answer. The router then picks from the models your workspace allows, with a fallback ready if a provider is slow.
Bring your own
Enterprise teams can add private deployments or their own keys. The router treats them like any other model, with the same logging and controls.
What we measure
We track answer quality, latency and cost per workspace, and share a monthly summary with admins so the trade-offs are always visible.


