Furiwake, Eval,
and Agents.

Furiwake is the routing-and-evaluation primitive for agent labor. It classifies every agent task — coding, research, customer-facing work, operations, analysis — and dispatches it to the model that gets the job right at the lowest cost. Routing decisions account for per-tenant policy, cost caps, latency budgets, and live signals from your evaluation set.

Our distinguishing claim isn't speed — it's that the policy stays calibrated as your traffic moves. Drift detection catches silent quality regressions before users do, and the fallback ladder is something your security team can actually audit.

Evaluation built around the principle that synthetic benchmarks lie about what you actually ship. Eval runs on real production traffic distributions, attributes regressions to the change that caused them, and gives you a regression budget instead of a green/red light.

Built for teams who want to release weekly without becoming superstitious about which prompt changes are safe.

Long-horizon agents for operationally complex work. Human approvals where they matter — high cost, high risk, irreversible — and autonomy everywhere they don't. Every decision leaves a trace you can replay, audit, and roll back.

Designed for the work where the model running unattended is exactly what you want, and also exactly what you fear. We make the second part tractable.