AutoR is a human-centered research operating system designed to transform complex, long-running research tasks into reproducible, artifact-backed runs. It targets researchers and engineers who need a robust execution system that maintains human oversight while leveraging AI for task completion. AutoR enhances research productivity by ensuring every step is auditable and outputs are verifiable, moving beyond mere documentation to tangible, inspectable results.

How It Works

AutoR employs a structured, 8-stage pipeline where a coding agent handles execution under the direct control of the human user. Crucially, human approval is mandatory after each stage, ensuring direction and quality. The system prioritizes verifiable outputs, generating a reproducible research artifact for each run stored on disk. This includes prompts, logs, code, data, figures, and writing sources, making the entire research process inspectable and auditable.

Quick Start & Requirements

• Prerequisites: Python 3.10+ and Claude CLI installed and available on the PATH. Local TeX tools are beneficial for manuscript compilation.
• Optional: google-genai library and a Gemini API key (or configs/diagram_config.yaml) are required for generating research diagrams.
• Execution: Initiate runs via python main.py, with options for specifying goals, preloading resources, resuming, redoing stages, selecting Claude models, and targeting specific publication venues.
• Links: A showcase example run is available at runs/20260330_101222.

Highlighted Details