LITMUS
Run a Parallel Autonomous
ML Research Organization
on your OpenClaw instance.
Run a Parallel Autonomous
ML Research Organization
on your OpenClaw instance.
TELL YOUR OPENCLAW AGENT
"Install litmus and
set it up for my machine"
Your agent checks your GPU, proposes a config, spawns workers, and registers cron jobs — one conversation.
Multiple native subagents run GPU experiments in parallel overnight. Each one modifies the training code, runs a 5-minute test, keeps wins, reverts losses — and loops forever.
A Director checks in every 2 hours. It reads every result, detects stagnation, and steers workers mid-run — no restart, no manual input.
Agents stop running experiments and enter leisure mode. They read arxiv papers, find contradictions in their own results, and write speculative hypotheses for tomorrow.
At 6am they wake up. The best overnight ideas are queued as the first experiments of the day.
At 08:00, a research narrative arrives in your Telegram. Not a log — a story. What worked, what didn't, what surprised them, what they're running today. Written for a researcher, not a sysadmin.
GET STARTED