Pricing
All features. All tiers. No credit card. BrainstormRouter is a free research project — everything is available at no cost while we build and test in the open. Planned GA tiers shown below.
These aren't upsells. They're the foundation.
You bring your own API keys. You pay providers directly at their published rates. BrainstormRouter never marks up model costs.
Works with any OpenAI SDK, LangChain, Vercel AI SDK, CrewAI, LlamaIndex. Change your base URL. That's it.
Adaptive routing is available on all plans. The router learns which models work best for your workloads from day one.
Vector similarity caching is included. Free tier gets 100MB; Pro gets 10GB; Enterprise is unlimited.
Guardian Intelligence headers ship on every response. You always know what a request cost, which model handled it, and why.
Automatic provider failover. If a provider goes down, your requests route to alternatives without code changes.
No. You bring your own API keys and pay providers directly. BrainstormRouter charges for the routing and intelligence layer only.
Yes. BrainstormRouter ships as a Docker image and npm package. Self-hosted deployments get all features. Enterprise customers get dedicated infrastructure support.
One API call to /v1/chat/completions or any other completion endpoint.
Cache hits count as requests (but at a lower rate). Memory operations are separate.
During the research phase, all tiers are free. Limits are soft — if you hit them, reach out and we'll raise them.
Yes. Cache hits don't call external providers, so you pay zero model costs. They count as BrainstormRouter requests at a reduced rate. Higher cache hit rates directly reduce your total spend.
No. Everything is free during the research phase. No credit card. No contracts.
Start free
See Thompson Sampling learn your workload. Watch costs drop in the Guardian headers.