From zero to a
signed run.
Four steps from a fresh account to evidence you hold. Add your own model key, point an agent at a real task inside the no-route sandbox, watch every call get policy-checked at the broker, then export the audit chain and re-verify it on your own machine.
Add a key. Get back proof.
Each step is the same on day one as it is on day one hundred. No infrastructure to stand up, no agent rewrite — the governance lives below your code.
Sign in & add your key BYOK
Sign in and open the API Keys page. Paste a provider key — bring your own from any supported provider — and it's sealed in an encrypted vault, never logged. The key stays with the broker and never enters a sandbox. You pay your provider for tokens; we never mark that up.
Start a run SANDBOXED
Pick an agent and give it a task. The run spins up in its own Docker container on an internal-only network with no route out — bounded by a time limit, a command limit, and a hard cost cap, and killable at any moment. Nothing about your task can open an outbound connection on its own.
Watch it work GOVERNED
The agent runs the task while every model call is routed through the one broker and policy-checked before it leaves — known model, budget not exceeded, kill-switch clear. You see spend against the cap in real time, and one click SIGKILLs the run from outside the sandbox if you want it stopped.
Export & re-verify OFFLINE
When the run finishes, export the audit chain — every prompt, tool call, output, routing decision, and kill, Ed25519-signed and hash-chained. Run the standalone verifier on your own machine and watch it pass. Tamper a single byte and verification fails. The proof is yours, not a screenshot.
Three things up front.
The answers to what people ask in their first ten minutes.
Supported models
Bring your own Anthropic or OpenAI key, or any OpenAI-compatible endpoint. The broker resolves the model and injects your key at call time — you keep control of the provider relationship and the bill.
BYOK · ANY KEYWhat a run is
One governed agent task, bounded by a time limit, a command limit, and a hard cost cap. A single run can make many model calls and use many tools — you are never billed per call by us.
ONE · TASKRe-verify anywhere
The exported chain re-verifies offline with the standalone verifier — no account, no network, no trust in us required. Hand it to an auditor and they can check every signature themselves.
NO · ACCOUNT NEEDEDYou don’t have to trust the run. Re-verify it. The export checks on your own machine, with our open tool — and if a single byte moved, it fails.
Run your first agent
and prove it.
Sign in, add a key, and point an agent at a real task — then export the signed proof and re-verify it yourself. Solo $29 · Team $544 · Business $1,779.