Documentation

The Python SDK inside your training process publishes live telemetry and safe points. The LaunchHelm dashboard at launchhelm.com charts the run, applies your controls, and keeps the record.

Start here

• Getting started — account, API key, first logged run.
• Python — install, configure, log, read back.
• Concepts — runs, participants, sessions, safe points, command outcomes.

Guides

• PyTorch — launchhelm.torch on an ordinary loop.
• Distributed training — DDP under torchrun.
• Control and tunables — built-in controls and custom commands.
• Checkpoints and resume — save, restore, crash recovery.
• Artifacts — files attached to a run.
• Alerts and notifications — rules and channels.
• The assistant — ask about your runs, with charts you can add to a board.
• Hugging Face and Lightning — callbacks for framework-owned loops.
• MCP — connect an agent client with a user key.

Reference

• Examples — runnable files to download.
• Limits — the numbers you can hit.
• Troubleshooting — errors and their fixes.

TypeScript, Rust and C++ SDKs are coming soon.