Awesome RAG (Retrieval-Augmented Generation)

A curated registry of open-source tools, frameworks, models, and datasets for building Retrieval-Augmented Generation (RAG) systems.

The goal is to be a practical starting point for assembling production-grade RAG pipelines: frameworks, vector stores, embedding models, document parsers, evaluators, graph and multimodal retrieval, inference engines, and benchmarks.

Inclusion criteria: Resources must be directly useful for RAG pipelines — retrieval, indexing, embedding, parsing, evaluation, deployment, or benchmarking. General LLM chat apps or pure research papers are out of scope. See CONTRIBUTING.md for the full quality bar.

---

Contents

• RAG Frameworks
• Vector Databases
• Embedding Models
• Document Loaders & Parsers
• RAG Evaluation
• Graph RAG
• Multi-Modal RAG
• RAG Deployment & Inference
• Benchmarks & Datasets
• Related Awesome Lists
• Contributing
• License

---

RAG Frameworks

End-to-end frameworks for building, orchestrating, and deploying RAG applications.

• LangChain Official — Industry-standard framework with document loaders, retrievers, chains, and agents.
  • Install: pip install langchain
• LlamaIndex Official — Data framework for connecting custom data sources to LLMs with advanced indexing and retrieval.
  • Install: pip install llama-index
• Haystack Official — Modular AI orchestration framework for production RAG, search, and agent pipelines.
  • Install: pip install haystack-ai
• DSPy Official — Stanford NLP framework for programming language models with declarative, retrieval-aware modules.
  • Install: pip install dspy
• RAGFlow Official — Open-source RAG engine focused on deep document understanding and citation-backed answers.
  • Install: docker compose -f docker/docker-compose.yml up -d
• Cognita Official — Modular, production-ready RAG framework with pluggable components.
  • Install: clone repo, then pip install -r requirements.txt
• Verba Official — Open-source, out-of-the-box RAG application powered by Weaviate.
  • Install: pip install goldenverba

Vector Databases

Vector and hybrid search engines that power the retrieval layer of RAG systems.

• Chroma Official — Local-first, AI-native embedding database.
  • Install: pip install chromadb
• Qdrant Official — High-performance, Rust-based vector database with rich metadata filtering.
  • Run: docker run -p 6333:6333 qdrant/qdrant
• Weaviate Official — Cloud-native vector database with hybrid (BM25 + vector) search and built-in reranking.
  • Run: docker compose up -d via the quickstart
• Milvus Official — Distributed vector database designed for billion-scale ANN search.
  • Run: docker run -p 19530:19530 milvusdb/milvus:latest
• Pinecone Official — Fully managed vector database with serverless indexing.
• pgvector Official — PostgreSQL extension for vector similarity search.
  • Install: CREATE EXTENSION vector; in Postgres

Embedding Models

Models and libraries for encoding text (and increasingly other modalities) into dense vectors for retrieval.

• Sentence Transformers Official — Standard Python library for training and running embedding and reranker models.
  • Install: pip install sentence-transformers
• all-MiniLM-L6-v2 Official — Fast, lightweight general-purpose sentence embedding model (384 dims).
• BGE-M3 Official — State-of-the-art multilingual embedding model supporting dense, sparse, and multi-vector retrieval.
• Nomic Embed Text v1.5 Official — Fully open-source (Apache 2.0) 768-dim embedding model with Matryoshka support.
• multilingual-e5-large Official — Microsoft E5 model trained for multilingual symmetric and asymmetric retrieval.
• Jina Embeddings v3 Official — Multilingual embedding model with task-specific instruction prefixes.

Document Loaders & Parsers

Tools that extract clean, structured text from PDFs, web pages, office documents, and other sources.

• Unstructured Official — Platform for turning complex documents into LLM-ready chunks with semantic element types.
  • Install: pip install unstructured
• LlamaParse Official — GenAI-native document parser for complex PDFs, tables, and charts.
  • Install: pip install llama-cloud-services
• Docling Official — IBM Research open-source toolkit for parsing PDFs, DOCX, and more into structured formats.
  • Install: pip install docling
• Marker Official — GPU-accelerated PDF-to-Markdown converter with layout preservation.
  • Install: pip install marker-pdf
• Firecrawl Official — API-first web scraping and document parsing service that outputs LLM-ready Markdown.
  • Install: npm install -g firecrawl or use the cloud API
• PyMuPDF4LLM Official — Fast PDF-to-Markdown extraction built on PyMuPDF for RAG pipelines.
  • Install: pip install pymupdf4llm

RAG Evaluation

Frameworks and platforms for measuring retrieval quality, answer faithfulness, and end-to-end RAG performance.

• RAGAS Official — Reference-free evaluation framework for RAG pipelines (faithfulness, relevancy, recall, etc.).
  • Install: pip install ragas
• ARES Official — Automated RAG evaluation system with synthetic data generation and confidence intervals.
  • Install: pip install ares-ai
• DeepEval Official — Pytest-like LLM evaluation framework with RAG-specific metrics.
  • Install: pip install deepeval
• TruLens Official — Open-source library for evaluating and tracking LLM app quality, including RAG.
  • Install: pip install trulens
• MLflow Official — Open-source AI engineering platform with RAG evaluation, tracing, and monitoring.
  • Install: pip install mlflow
• Arize Phoenix Official — LLM observability and evaluation platform with RAG-specific judges.
  • Install: pip install arize-phoenix

Graph RAG

Tools that combine knowledge graphs with retrieval to improve multi-hop reasoning and structured grounding.

• Microsoft GraphRAG Official — Modular graph-based RAG system using LLM-derived knowledge graphs and community summaries.
  • Install: pip install graphrag
• Neo4j GraphRAG Python Official — First-party Neo4j package for building GraphRAG applications in Python.
  • Install: pip install neo4j-graphrag
• LightRAG Community — Simple and fast graph-based retrieval-augmented generation.
  • Install: pip install lightrag-hku
• MsGraphRAG-Neo4j Community — Neo4j implementation of the Microsoft GraphRAG methodology.
  • Install: pip install ms-graphrag-neo4j
• Neo4j Official — Native graph database with vector indexes and GraphRAG integrations.
  • Run: docker run -p 7474:7474 -p 7687:7687 neo4j:latest
• LLMGraphTransformer Official — LangChain utility for building knowledge graphs from unstructured text.

Multi-Modal RAG

Retrieval systems that handle images, documents-as-images, video frames, and other non-text modalities.

• ColPali Community — Visual document retrieval using vision-language models and late-interaction embeddings.
  • Install: pip install colpali-engine
• ColQwen2 Official — ColPali-style model based on Qwen2-VL for visual document retrieval.
• Byaldi Community — Simple wrapper around ColPali/ColQwen models for multimodal RAG.
  • Install: pip install byaldi
• CLIP Official — OpenAI's contrastive language-image pre-training model for cross-modal retrieval.
  • Install: pip install git+https://github.com/openai/CLIP.git
• SigLIP Official — Google's sigmoid loss for language-image pre-training, used in multimodal retrieval.

RAG Deployment & Inference

Inference engines and serving platforms for running the generation (and sometimes embedding) side of RAG.

• vLLM Official — High-throughput LLM inference engine with PagedAttention and OpenAI-compatible serving.
  • Install: pip install vllm
• Ollama Official — Local, easy-to-use LLM runner with built-in quantized model support.
  • Install: curl -fsSL https://ollama.com/install.sh | sh
• Text Generation Inference (TGI) Official — Production LLM serving toolkit from Hugging Face.
  • Run: docker run ghcr.io/huggingface/text-generation-inference:latest --model-id <model>
• SGLang Official — Fast serving framework for LLMs and reasoning models.
  • Install: pip install sglang
• BentoML Official — Unified model-serving framework with RAG and vLLM deployment examples.
  • Install: pip install bentoml

Benchmarks & Datasets

Standard datasets and leaderboards for evaluating retrieval and end-to-end RAG quality.

• MTEB Official — Massive Text Embedding Benchmark covering retrieval, classification, clustering, and more.
  • Install: pip install mteb
• BEIR Official — Heterogeneous benchmark for zero-shot information retrieval across 18 datasets.
  • Install: pip install beir
• Natural Questions Official — Real Google search queries with Wikipedia answers for open-domain QA.
• MS MARCO Official — Large-scale passage-ranking dataset based on real Bing search queries.
• HotpotQA Official — Multi-hop question answering benchmark requiring reasoning across multiple Wikipedia documents.
• RAGTruth Community — Dataset for studying hallucinations in RAG-generated responses.
• FRAMES Official — Factuality, retrieval, and reasoning benchmark for multi-hop RAG evaluation.

Related Awesome Lists

• Awesome Vector Databases — Curated list of vector search engines, embeddings, and ANN libraries.
• Awesome MCP Servers — Registry of Model Context Protocol (MCP) servers that can augment RAG agents.
• Awesome Open Source LLMs — List of open-source large language models suitable for RAG generation.

Notes

• Official vs Community: Official entries are maintained by the vendor or project owner; Community entries are third-party implementations.
• Managed services: Some entries (e.g., Pinecone, LlamaParse, Firecrawl) are primarily hosted APIs with local SDKs; see each project's docs for authentication and pricing.
• Evolving ecosystem: The RAG tooling landscape changes rapidly. If you spot a broken link or a missing high-quality project, see CONTRIBUTING.md.

Contributing

Read CONTRIBUTING.md for the quality bar, entry format, and PR process.

License

This list is released into the public domain under CC0-1.0.

Want us to build this for you?

Enterprise AI Atlas is maintained by Vibe Coding Agency. We prototype and ship agentic systems, MCP servers, and enterprise AI integrations for teams that need working software fast — without hiring a full AI engineering team.

Free guide: The Non-Technical Founder's Guide to Agentic AI — what agents and MCP servers are, and how to get a system built.