Caveman Router · coming soon
The world's firstcache-aware model router.
Every request runs on the cheapest model that can do the job — and your cache never breaks. Same quality. 39.6% less.
- 39.6%
- lower spend
- 92%
- pass rate, held
- +5 ms
- added, p50
nine weeks of replayed production traffic · modeled at list prices
- 01The waste
Most teams send every request to the biggest model. Easy work gets billed like the hardest work of the day.
- 02The reinvention
Per-call routers chase list price and wreck your cache. We rebuilt routing from the ground up to price the whole session.
- 03The gate
Nothing moves until it wins on your own evals — record, replay, shadow, canary. Slip once and it rolls itself back.
Watch the bill fall.
Nine weeks of real traffic, three ways in. Flip the switch, play the session, read the bench.
02 · Plug in
One line. Any SDK. Any coding agent.
Router speaks the OpenAI-compatible API. Point your base URL at it and keep everything else exactly as it is.
baseURL: "https://router.caveman.so/v1"Stop paying flagship prices for routine requests.
in private development · we'll reach out when there's something to point your base URL at