Agent Lab
Start hereSetupRecipesTargetsCreate an account

You run these

Recipes

Each one clones from a repository and runs on your own machine against the Lab's hosted infrastructure. Each says what it teaches, what it needs first, and roughly how long a first run takes.

introductory

Hello, GatewayClone a repo, run lesson one, and watch a local call reach a hosted server through the Agent Gateway.Planned30 mingateway

intermediate

Bring the agent you already haveTake an agent you have already built and route it through the Gateway, changing as little as possible.Planned45 mingatewayLet an agent act for you without giving it your credentialsYour agent calls an API on your behalf and never holds the secret — the Gateway brokers it, scoped to you.Planned60 mingatewayTry to make an agent misbehavePut guardrails around an agent's inputs and outputs, then spend an hour genuinely attempting to defeat them.Planned60 minstream, gateway

advanced

Two organisations, two agents, one conversationHave your agent discover and work with an agent belonging to somebody else — across a boundary, with the trust decision enforced and visible.In development120 mingatewayLet an agent spend moneyA mock-commerce flow where an agent pays for what it calls — with a limit it cannot exceed, on a wallet you control.Planned90 minpay, gateway

Status is not decoration. Runnable now means you can clone it today and it works. Planned means the shape is agreed and the work is not done. Nothing is listed as runnable on the strength of an intention.

Operated by MISSION. An experimental environment — no warranty, no SLA, and not a place for real or sensitive data.

Start here