autoresearch-agent

What problem does it solve?

The Autoresearch Agent automates an autonomous experiment loop that edits a single target file, evaluates changes with a fixed metric, and iterates toward improvement while handling git-based commits and rollbacks.

Core Features & Use Cases

• One-change-per-iteration: updates only the targeted file to isolate effect of each change.
• Deterministic evaluation loops: runs a defined evaluator to produce a measurable metric.
• Practical workflows: speeds up code optimization, performance tuning, and content improvements by repeated experimentation in a project.
• Real-world scenario: a team improves a critical module's latency by small, measurable increments over many iterations.

Quick Start

Start by creating a new experiment with the setup tool, then run a single iteration to begin optimizing your target file.