AutoResearchFrom Idea to Paper.

Multi-agent research orchestration coordinating Claude Opus, Codex, and Gemini for automated deep learning research.

🧠

Six Specialized Agents

Orchestrator, Planner, Writer, Scout, Coder, Judge — each with clear boundaries and the right LLM.

🔄

Work is always reviewed by a different LLM. Claude reviews Codex code, Codex judges Claude's experiment results.

📡

Active watch during training start, CronCreate periodic patrol after stabilization. Health-based, not timeout-based.

🎛️

Control automation level per stage: human approval, auto-judge, or fully automatic. Change at runtime with /gate.

📂

All research state persists in .omc/research/. Context is working memory, disk is long-term memory. Sessions can restart seamlessly.

🔀

autopilot (between stages), ralph (within-stage loops), ultrawork (parallel execution). Combine freely.