Summary

FAROS is a blueprint-driven AutoResearch runtime designed to automate research workflows, moving beyond single-agent AI scientist applications. It targets researchers and engineers seeking extensible, configurable automation, offering a structured approach to transform ideas into experiments, papers, and reviews, starting with the LLM domain.

How It Works

FAROS operates on a layered architecture: Blueprints define workflow graphs and validation, Capabilities implement research steps (e.g., idea refinement, paper drafting), Profiles bind blueprints to execution strategies, and Providers supply underlying engines like LLMs or tools. This release focuses on a runtime boundary separating core orchestration logic from reusable domain modules, enabling LLM as just one provider type for future cross-domain extensibility. The current workflow chain is idea -> experiment -> paper -> review.

Quick Start & Requirements

• Backend Install: Navigate to backend/, run pip install -r requirements.txt.
• Frontend Install: Navigate to frontend/, run npm install and npm run dev.
• Prerequisites: Python 3.11+ or 3.12, Node.js 18+, latexmk and pdflatex for PDF compilation, and a configured LLM provider.
• Documentation: Developer Guide (docs/DEVELOPER_GUIDE.md), TODOs (docs/FAROS_TODO.md).

Highlighted Details

• Supports venue-aware LaTeX paper generation using templates for ICML, NeurIPS, ICLR, ACL, and a generic fallback.
• Implements a complete LLM research workflow: idea -> experiment -> paper -> review.
• Separates core FAROS runtime (backend/app/faros/) from domain-specific modules (backend/app/modules/).
• Offers a "plan-only" execution mode via API for workflow simulation.

Maintenance & Community

No specific details on maintainers, community channels (e.g., Discord, Slack), or roadmaps were found in the provided README excerpt. The docs/FAROS_TODO.md file outlines future development priorities.

Licensing & Compatibility

The README excerpt does not specify the project's license or provide compatibility notes for commercial use.

Limitations & Caveats

This release candidate is a baseline for LLM-domain AutoResearch, not the final cross-domain platform. Key features not yet included are full DAG scheduling, generalized non-LLM providers, a complete experiment execution loop, a dedicated frontend console, and DB-backed metadata. Future work includes replacing the experiment scaffold with code synthesis/execution and enhancing verification.