documentation

Everything you need to ship your model.

From your first compile run to a live endpoint — guides, references, and examples for every step.

Quick start

Compile your agent

Point train.cloud at your running agentic system. It captures the tools, prompts, and decision routes into a portable training spec.

Run training rounds

Set a budget cap, kick off training, and watch the loss curve stream in real time. Each round generates a fresh eval scorecard.

Promote past the gates

Nothing reaches your endpoint until it passes every gate. Promote a version that clears quality, latency, safety, and cost checks.

Call your endpoint

Swap one base URL. Your existing client talks to the model you compiled — versioned, gated, and compatible with any OpenAI-compatible SDK.

Reference

Endpoint API

Chat completions, models, and embeddings. Drop-in OpenAI-compatible.

Coming soon

CLI reference

All compile, run, serve, and key management commands.

Coming soon

Eval gates

How the five promotion gates work and how to tune thresholds.

Coming soon

Credits and billing

Metering, caps, alerts, and top-up.

Coming soon

Ready to start?

Sign up free — your first 100 credits are on us.

Create account