RAG・知識ベース
gno
Local AI-driven document search and editing with first-in-class hybrid retrieval
★ 115 スターMIT
ガイドを見る →VideoRAGは、HKUDSが公開しているオープンソースのRAG・知識ベースプロジェクトです（GitHub: HKUDS/VideoRAG）。GitHubスターは3,389、フォークは476で、NOASSERTIONライセンスの下で公開されています。GitHubでの最終更新は2026年3月18日です。
VideoRAG — chat with your videos, even hundreds of hours of them, with graph-driven retrieval.
Decide the first concrete outcome you want from it, so you can pick the right example, template or mode instead of exploring everything at once.
Clone the repository (or install the published package) and skim its README quick start for the requirements for your platform.
Run the documented install step and set any required configuration — such as provider keys, data sources or local services — before the first run.
Execute the smallest documented example end to end, confirm the output is what you expected, then adapt it to your real data or workflow.
Multi-modal knowledge graphs structure video understanding instead of flat frame sampling.
Process videos up to hundreds of hours with hierarchical context encoding.
Compare and analyse multiple videos simultaneously with cross-video semantic modelling.
Text queries are aligned with visual and audio content to find specific moments precisely.
A bundled benchmark of 160+ videos / 134+ hours across lectures, documentaries and entertainment.
Designed to handle very long content on a single RTX 3090 (24GB).
Get the source so you can inspect, configure and run it locally.
git clone https://github.com/HKUDS/VideoRAG.git
cd VideoRAGClone the repo and follow the VideoRAG-algorithm setup: conda environment, checkpoints download and dependency installation.
git clone https://github.com/HKUDS/VideoRAG.git
cd VideoRAG # see VideoRAG-algorithm for conda setup and checkpoint downloadsFollow the Vimo-desktop instructions in the repo to run the desktop experience from source.
# see the Vimo-desktop folder in the repo for its install stepsIt is maintained by HKUDS on GitHub (HKUDS/VideoRAG), where the source, issues and releases live.
No clear licence was identified from the repo metadata or README, so check the repository's licence file before relying on it commercially.
The framework is open research code; running it uses vision/LLM models as configured in its setup (local checkpoints or provider APIs), so costs depend on that choice.
Compared with frame-sampling video chat demos or generic multimodal RAG, VideoRAG's distinguishing focus is: videoRAG — chat with your videos, even hundreds of hours of them, with graph-driven retrieval. Pick it when that focus matches your need; the alternative may suit you better if its broader or different scope is the priority.
Clone https://github.com/HKUDS/VideoRAG (or install its published package), follow the README quick start to install and configure it, then run the smallest documented example — the Run section on this page mirrors those steps.
Local AI-driven document search and editing with first-in-class hybrid retrieval
Build Neo4j knowledge graphs from unstructured data using LLMs — then chat with your graph.
Transparent Python RAG reference with FAISS + BM25 hybrid retrieval.
Self-hosted NotebookLM alternative — chat with documents, get audio summaries.
An agentic AI data assistant - chat with your databases, run SQL and code, get reports
Unstract — LLM-driven extraction of unstructured data, built for API deployments and ETL pipeline workflows.
LibreAIカタログに掲載されたRAG・知識ベースプロジェクトの中で最もスターの多いもの — VideoRAGが合わなかった場合に検討したい、現実的な候補リストです。
Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or…
RAGFlow is a leading open-source Retrieval-Augmented Generation (RAG) engine that fuses cutting-edge RAG with Agent capabilities to create a superior context…
Lightweight OCR toolkit turning PDFs and images into structured, LLM-ready data in 100+ languages
MinerU — turns PDFs and Office documents into LLM-ready Markdown and JSON for agents and RAG pipelines, Apache-2.0 plus terms
Stop renting your intelligence. Own it with AnythingLLM. Everything you need for a powerful local-first agent experience
カタログの中でVideoRAGとスター数が近いRAG・知識ベースプロジェクト — 前後の棚もそのまま見て回れます。
A personal knowledge base that builds itself — drop in sources, get a persistent interlinked wiki.
Fast, lightweight Python embeddings from Qdrant — dense, sparse, ColBERT, image and rerankers.
Knowhere turns documents into structured, hierarchy-native chunks ready for agents and RAG.
seekdb — the AI-native search database unifying vector, text and structured data for agents.
The open-source, permission-aware context layer that lets agents search all your company knowledge.
Federated AI search and RAG across 100+ apps — answers with sources, without moving your data