Any research. One Claw. 🦞
Overview
v0.1.0 is the first official release of OneResearchClaw.
OneResearchClaw is a skill based end-to-end research report generation pipeline, supporting automatic content structuring (Grounding), literature research (Literature Research), quality review (Review), and multi-format export (Export) from various sources including meeting recordings/videos, documents, spreadsheets, PPT files, ZIP mixed material packages, and arXiv/YouTube/Bilibili links.
✨ Core Highlights
| Core Capability | Description |
|---|---|
| 🧭 Unified Multi-source Input | Supports many types of materials, from meeting recordings to PDFs, from tables to ZIP packages. The system automatically identifies the material type and routes it to the corresponding grounding skill, reducing manual organization, conversion, and preprocessing costs. |
| 🧠 Multi-Topic Meeting Splitting | For long meetings, complex discussions, or multi-issue materials, the system can automatically split content into multiple topics and generate independent research branches. Each topic can separately go through grounding, research, summary, and export, making it suitable for producing multiple focused reports from one complex meeting. |
| 🎚️ Controllable Research Depth | Supports simple / medium / complex research modes to control literature search, source opening, evidence organization, and report analysis depth. It can generate a quick briefing or a more complete research report. |
| 🔎 Query Confirmation & Directed Search | Before formal retrieval, the system first generates candidate queries based on the grounded note and supports user confirmation, addition, deletion, or modification. This makes literature search better aligned with the user's real research direction and reduces irrelevant retrieval and token waste. |
| 🧪 Multi-model Review Mechanism | After report generation, the report enters a review → rewrite loop. Different models or review perspectives can check completeness, evidence consistency, logical coherence, and over-claiming, then revise the report through bounded rounds. |
| ☁️ One-click Cloud Link Ingestion | Supports cloud link ingestion and processing for sources such as arXiv, YouTube, and Bilibili, allowing online materials to enter the same research pipeline directly. |
| 📦 Multi-format Export | Supports exporting final results as md / docx / pdf / pptx / audio, covering reading, archiving, presentation, and reuse scenarios, with both Chinese and English export supported. |
| 🧬 Skill-Evolve | Supports evolving skill subsets based on user feedback through feedback collection, patch generation, regression checks, and version promotion, so the system can continue improving rather than staying fixed as a one-off hand-written workflow. |
Installation
# Core functionality only
pip install -e .
# Audio processing (meeting audio/video)
pip install -e ".[audio]"
# Document parsing
pip install -e ".[document]"
# Report export
pip install -e ".[export]"
# Spreadsheet processing
pip install -e ".[table]"
# External API (literature search)
pip install -e ".[api]"
# Full installation
pip install -e ".[full]"Python version requirement: 3.10+
Technical Stack
- Speech Recognition: Faster-Whisper Large v3 (local inference)
- Speaker Diarization: pyannote Speaker Diarization 3.1
- Document Parsing: Docling
- Model Management: Built-in
one-report-prepare-modelsCLI tool - Build Tool: hatchling
License: MIT
Repository: https://github.com/gaotiexinqu/OneResearchClaw
Page: https://github.com/Weiyang209/any2one