AI Research OS¶

A Self-Evolving Research Operating System for AI Researchers · Built with MkDocs

AI Research OS is a local-first research tool that grows smarter over time. It learns your research patterns, surfaces what matters, and generates insights from your paper library.

What It Does¶

Feed it a paper — get back structured, cross-linked research knowledge:

Core Philosophy¶

Not a PDF manager. A self-evolving research partner that:

• Learns from your research patterns
• Improves answers over time
• Adapts to your specific domain
• Surfaces gaps and opportunities

Quick Start¶

pip install -e ".[all]"

# Import a paper
python -m cli import 2601.00155 --tags LLM,Agent

# Search your library
python -m cli search "attention mechanism" --tag LLM

# Autonomous research
python -m cli research "RLHF alignment" --limit 5

# Chat with your papers
python -m cli chat-tui

See Installation for full setup instructions.

Key Features¶

Key New Capabilities¶

• Adaptive Scheduling — GenePool saturation-aware research interval
• Gap Clustering — Semantic clustering of research gaps with hotspot trend analysis
• Contradiction Timeline — Detect paradigm shifts from conflicting paper claims
• Impact Tracking — Quantified research impact: novelty × depth × strength × speed
• Parallel Research — Multi-agent concurrent gap analysis with result merging
• Topic Discovery — Gap-density-based intelligent subscription suggestions
• Rich Webhooks — Discord embeds + Feishu cards for gap and paradigm shift alerts
• Structured Observability — JSON logging with correlation IDs and event tracking

23 CLI Commands¶

• import — Bulk import from arXiv, DOI, PDF
• search — Full-text search with BM25 ranking
• chat-tui — Full-screen TUI chat with paper context
• kg — Knowledge graph query and rebuild
• gap — Detect research gaps, generate research questions
• rag — RAG pipeline: paper → code → tests → benchmark
• benchmark — Cross-paper benchmark with D3.js charts
• paper2code — Generate code from paper
• subscribe — RSS-style paper feed by tag/query

Research Knowledge Structure¶

Paper → P-Note (per paper)
      → C-Note (per concept/tag)
      → M-Note (comparison when 3+ papers share a tag)
      → Radar (topic frequency heat score)
      → Timeline (year-based evolution)

Integrations¶

• arXiv — Direct import by ID or URL
• OpenAlex — Citation graph (forward + backward)
• Ollama — Local embeddings (nomic-embed-text, 768-dim)
• DashScope / OpenAI — AI draft generation
• EvoSkill — Benchmark-driven skill discovery
• Streamlit — Optional web dashboard

Project Status¶

Resources¶

• Usage Guide — Full command reference
• Architecture — System design
• Configuration — Environment variables
• Contributing — How to contribute
• Roadmap — Where we're going
• GitHub

License¶

GNU General Public License v3.0 — see LICENSE for details.