← all repositories
EvoMap/AutoResearch

An Agent Workflow That Treats Negative Results as Evidence

AutoResearch automates the full AI/ML research lifecycle—from literature mining and idea generation through experiment execution and multi-model peer review—while keeping every plan, failure, and log on disk for human audit.

AutoResearch
Collecting fresh signals — velocity needs a few days of history.
collecting data…
star history

What it does

AutoResearch is a stateful agent workflow that carries an AI/ML research idea from initial hunch to paper-ready evidence. It discovers directions by cross-referencing recent papers, community discussions, and open-source trends against a local knowledge base, then plans experiments, executes them, and subjects results to independent multi-model review. Every artifact—plans, code, run logs, metrics, failure causes, critic reports, and blind reviews—is written to disk so researchers can inspect, take over, or halt the process.

The interesting bit

Most agent frameworks celebrate successes; this one treats negative results as evidence worth preserving and stops a failed hypothesis instead of forcing a success story. It also enforces intellectual hygiene by requiring at least three distinct models to cross-review ideas during generation, so no single model gets to hallucinate and grade its own homework.

Key highlights

  • Cross-domain idea generation that intersects external research signals with local domain knowledge constraints
  • Stateful, recoverable execution that persists plans, code, queues, and logs so long runs survive interruption
  • Independent multi-model review stages that use distinct underlying model identities to check important claims
  • Explicit support for negative results: preserves evidence and halts when a hypothesis fails
  • Traceable provenance records Forge sources, knowledge directions, experiment results, and critic reports

Caveats

  • The system explicitly warns that it cannot guarantee every conclusion is correct; it preserves evidence for researcher review rather than replacing human judgment.
  • When evidence is thin, research agents may still invent missing details, which is a known risk the workflow tries to mitigate but not fully eliminate.

Verdict Worth exploring if you run a research lab or want to automate literature-to-experiment pipelines with auditability. Skip it if you are looking for a guaranteed correct paper generator or a simple notebook tool.

Frequently asked

What is EvoMap/AutoResearch?
AutoResearch automates the full AI/ML research lifecycle—from literature mining and idea generation through experiment execution and multi-model peer review—while keeping every plan, failure, and log on disk for human audit.
Is AutoResearch open source?
Yes — EvoMap/AutoResearch is open source, released under the Apache-2.0 license.
What language is AutoResearch written in?
EvoMap/AutoResearch is primarily written in Python.
How popular is AutoResearch?
EvoMap/AutoResearch has 3k stars on GitHub.
Where can I find AutoResearch?
EvoMap/AutoResearch is on GitHub at https://github.com/EvoMap/AutoResearch.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.