Drag the sliders to match your usage. Estimates assume local-routed prompts cost effectively nothing after setup.
You don't have to take the calculator's word for it. The proxy measures the real number as you work: every request is tallied in a local ledger against frontier baseline pricing ($3 per million input tokens, $15 per million output). kultivait harvest shows exactly what you spent, what the everything-to-cloud default would have cost, and the difference.
The biggest lever. Most day-to-day prompts are routine, and routine prompts route to a model that costs you nothing per token.
When the cloud is needed, the router still picks the cheapest capable tier rather than defaulting to the flagship model for everything.
Once you see per-prompt costs in the harvest ledger, wasteful patterns get fixed. What gets measured gets optimized.
There's a fourth effect the calculator can't capture: context hygiene. kultivait prune distills long transcripts before they snowball, so the prompts you do send — local or cloud — carry fewer wasted tokens. The greenest token is the one you never send.
Kultivait.ai is free during early access — install it, run it, keep the savings. If paid plans come later, the principle stands: a fraction of what it saves you, so the math always works in your favor.
Install in one command, point your tools at the proxy, and watch kultivait harvest count your actual savings.
curl -fsSL https://kultivait.ai/install.sh | sh
Setup details and the full command reference live on the developers page.