Reference

Terminal (CLI)

The rakuvo command: run the gateway, benchmark a configuration, test an agent's tools.

text
rakuvo serve [flags] run the gateway
rakuvo bench [flags] run task sets with and without layers
rakuvo autotest [flags] generate and check test cases for an agent's tools

#serve

Starts the OpenAI-compatible gateway. See the configuration reference for every flag.

bash
rakuvo serve -listen 127.0.0.1:8081 -tool-policy-auto -loop-guard -trace ./traces.jsonl

Endpoints: POST /v1/chat/completions, POST /v1/responses, GET /healthz.

#bench

Runs a task set through several arms (a baseline and one or more layer configurations) and reports accuracy, cost and latency. Useful for checking a layer on your own tasks before you rely on it. Use -dry-run to see what would run and -max-cost to cap spend.

#autotest

Generates test cases for an agent's tool set with one model and cross-checks them with another, so you can find broken tool definitions before an agent does. A runner for push-time testing in CI is on the roadmap and is not built yet.

#Running a gateway for a team

bash
export RAKUVO_TOKEN=$(openssl rand -hex 24)
export RAKUVO_API_KEY=rk_...
rakuvo serve -listen 0.0.0.0:8081 -token-env RAKUVO_TOKEN -config-from-api -api-url https://your-rakuvo-api

The gateway now takes its layer settings from the control plane and picks up changes within one poll interval, with no restart.