Documentation

Everything you need to run AI models for a robot or a simulation. Start with the quickstart. AI agents can read every page as markdown: see llms.txt.

Start here

• QuickstartInstall the SDK, add your key, and send your first request to a model running on our GPUs, with no GPU of your own.
• Guide for AI agentsHow an AI coding agent should read these docs, get a key, handle errors, and check that a robot integration really works.

Reference

• Python SDK referenceEvery public class and method in robot-inference-client, the Python SDK, with the options that matter in a robot control loop.
• Command line (yb)The yb command for checking your key, listing models, running a test session, reading the docs and explaining errors from a terminal.
• Robot data formatsThe exact names, image sizes and position numbers each model expects from your robot, and the shape of the moves it sends back, taken from the code that checks them.
• ErrorsEvery error code and every reason a session can end, with what it means, whether to retry, and how to fix it.
• HTTP and WebSocket APICall the API directly, without the Python SDK. Authentication, errors, safe retries, every route, and the session stream.

Policies

• Billing and creditYou pay per request the model answers, from prepaid credit, capped by each session's max_spend_usd. What is charged, what is free, and how a session is settled.
• LimitsThe limits the service enforces today, such as message size, deadlines, open sessions and rate limits, and why each one exists.
• Security and dataWhat keys and session passes can do, how your data travels, what we keep and for how long, and how to report a security problem.
• Questions and answersShort answers about what Yellow and Black is, sessions, what you pay for, speed, supported robots, your data, and AI agents.