RAG

RAGFlow is an open-source RAG engine combining deep document understanding with agentic retrieval and workflow orchestration. It chunks heterogeneous files (PDF, Office, scans), supports configurable LLMs/embeddings, MCP, and produces grounded answers with traceable citations. Self-hostable via Docker.

RAG | Go · retrieval-augmented-generation · agentic-retrieval

Docling is a Python library and CLI that parses PDFs, Office files, HTML, EPUB, images and audio into a unified DoclingDocument, exporting to Markdown, HTML or lossless JSON. It is widely used as the document ingestion stage for RAG and agentic AI pipelines.

RAG | Python · document-parsing · pdf

Self-hosted, local-first AI app: ingest documents into a vector DB and chat with them privately, with built-in agents, no-code agent flows, MCP support, multi-user permissions and a wide range of local or cloud LLMs and embedders.

RAG | JavaScript · local-first · self-hosted

Local-first AI memory system that stores conversation and project history verbatim and retrieves it via pluggable vector backends (ChromaDB default, plus Milvus/Qdrant/pgvector/SQLite/Rust). Ships a CLI, MCP server, and agent skills; benchmarks 96.6% R@5 raw on LongMemEval with no API key.

RAG | Python · ai-memory · long-term-memory

Milvus is a cloud-native distributed vector database for scalable ANN search over embeddings. It supports HNSW, IVF, DiskANN, GPU indexes, BM25 full text and hybrid dense/sparse search, making it a common retrieval backend for RAG and multimodal AI apps.

RAG | Go · vector-database · ANN-search

LightRAG is a Python RAG framework that builds a dual-level knowledge graph over documents for entity- and relation-aware retrieval. It ships an API server, WebUI, multiple storage backends (Neo4j, Postgres, Mongo, Milvus, OpenSearch), reranking, multimodal parsing and OpenAI/Ollama/Gemini integrations.

RAG | Python · knowledge-graph · graphrag

Microsoft GraphRAG is a modular Python pipeline that uses LLMs to extract entities and relationships from unstructured text, build a knowledge graph, cluster communities, and answer queries with graph-based retrieval. Note: the repo is largely in maintenance mode with only bug fixes and CVE updates.

RAG | Python · graphrag · knowledge-graph

Python SDK for PageIndex, a vectorless RAG engine that builds a hierarchical tree index per document and lets an LLM reason over that tree to retrieve the right section. Supports local or cloud indexing, agent/MCP integration, and citations; no vector DB or chunking.

RAG | Python · vectorless-rag · reasoning-based-retrieval

Qdrant is a Rust-written, production-ready vector database and similarity search engine with REST/gRPC APIs, filtering, hybrid search, quantization, and sharding. It's a common retrieval backend for RAG, semantic search, recommendations, and image search.

RAG | Rust · vector-database · vector-search

Onyx is a self-hosted AI knowledge platform that indexes content from 50+ apps for permission-aware enterprise search and agentic RAG. It also provides AI agents, deep research, web search, MCP access, and integrations with hosted or local LLMs.

RAG | Python · agentic RAG · enterprise search

STORM researches topics through web retrieval and multi-perspective LLM conversations, then generates outlined reports with citations. Co-STORM adds collaborative human-AI discussion and a dynamic mind map.

RAG | Python · deep research · agentic RAG

Python framework for building temporal knowledge graphs that give AI agents time-aware memory. Ingests structured and unstructured episodes into Neo4j/FalkorDB and serves hybrid semantic+keyword+graph retrieval. Ships an MCP server.

RAG | Python · knowledge-graph · temporal-graph

Cognee is a self-hosted memory and retrieval platform that turns documents, code, and conversations into searchable knowledge graphs and vector context. Use its Python API, REST API, or MCP server to add persistent memory to agents.

RAG | Python · agent memory · long-term memory

Supermemory is a memory and context engine for AI: it extracts facts from conversations, maintains user profiles (~50ms), and serves hybrid RAG+memory search over one API. Self-hostable via one binary with local embeddings, plus MCP server and plugins for Claude Code, Cursor and other clients.

RAG | TypeScript · agent-memory · long-term-memory

Chroma is an open-source vector and search database for AI, providing a 4-function API to store documents, embeddings, and metadata and query by similarity. Ships Python/JS/Rust clients plus a dockerizable client-server mode for RAG pipelines.

RAG | Rust · vector-database · embeddings

WeKnora is a self-hostable, LLM-powered knowledge platform in Go: it ingests 10+ document formats into a RAG pipeline, adds a ReAct agent with MCP tools and sandboxes, and auto-generates a self-maintaining markdown wiki. Supports pluggable vector stores, 20+ LLM providers, multi-tenant RBAC, and IM channel integration.

RAG | Go · knowledge-base · document-qa

Postgres extension adding vector types plus exact and approximate nearest neighbor search with HNSW and IVFFlat indexes. Supports single, half, binary, and sparse vectors with L2, inner product, cosine, L1, Hamming, and Jaccard distances for embeddings and RAG retrieval.

RAG | C · vector-search · embeddings

Rust library that classifies PDFs (text-based, scanned, image, mixed) in milliseconds and extracts position-aware text, tables, and clean Markdown, with optional per-page OCR routing. Ships Python, Node, WASM bindings and CLI tools, aimed at fast local document ingestion for LLM/RAG pipelines.

RAG | Rust · pdf-parsing · text-extraction

turbovec is a Rust vector index with Python bindings built on Google's TurboQuant quantizer. It offers online ingest with no training step, SIMD-accelerated search reported faster than FAISS FastScan, allowlist-filtered search, and incremental crash-safe saves.

RAG | Rust · vector-search · ann

Weaviate is a Go-based, cloud-native vector database storing objects and vectors together, so you can combine vector similarity search with structured filtering, hybrid BM25 search, integrated vectorization, RAG, and reranking in one API.

RAG | Go · vector-database · semantic-search

A local-first NotebookLM alternative for indexing documents and answering questions with citations. It can generate study materials, reports, slides, and podcasts on your machine, with local models and air-gapped use supported.

RAG | Python · local-first · air-gapped

Python library that partitions 60+ document types (PDF, HTML, DOCX, email, images) into structured elements for LLM and RAG pipelines, with chunking, enrichment, and an MCP server for agent workflows.

RAG | HTML · document-parsing · etl

memU captures knowledge from agent sessions, stores it as searchable memory, and retrieves relevant context across agents and devices. It provides host adapters for coding agents, local SQLite or PostgreSQL storage, and automatic skill extraction.

RAG | Python · agent-memory · cross-agent-memory

EverOS is a local-first memory runtime for AI agents: memories are stored as plain Markdown with SQLite + LanceDB indexes for hybrid keyword/vector retrieval. It adds reflection, skill extraction, a Knowledge Wiki, and MCP/plugin integrations, requiring only Python 3.12 and one LLM API key to start.

RAG | Python · agent-memory · long-term-memory

LEANN is a local, privacy-first vector database for RAG that uses graph-based selective recomputation to cut embedding storage by ~97% versus traditional vector DBs, with no accuracy loss. It indexes documents, emails, browser history, chat logs, and code, and ships an MCP server for agent use.

RAG | Python · vector-search · vector-database

Open-source engine behind r.jina.ai and s.jina.ai: prepend a URL to get LLM-ready markdown from web pages, PDFs, Office files and images, or a search query to get the top results already fetched. Runs stateless or with S3/MinIO caching.

RAG | TypeScript · url-to-markdown · web-scraping

CocoIndex is a Python-facing, Rust-backed framework for declaratively building incremental data and RAG pipelines. It recomputes only data affected by source changes and supports continuously refreshed search indexes and agent context.

RAG | Rust · incremental indexing · agent context

PixelRAG renders web pages and PDFs to screenshot tiles and retrieves over the images with a LoRA-tuned Qwen3-VL embedding model, so tables, charts and layout survive retrieval. Includes a pixelshot CLI, FAISS/Qdrant indexing, a FastAPI search server, a hosted 8.28M-page Wikipedia index, and a Claude Code screenshot sk

RAG | Python · visual-rag · pixel-native-search

Rust + Postgres knowledge platform combining a bitemporal knowledge graph, GraphRAG hybrid search, ontology-driven reasoning and an MCP server so agents can read governed enterprise knowledge. Self-hosted, air-gappable, with review queues and an append-only decision ledger.

RAG | Rust · graphrag · knowledge-graph

Vespa is a distributed platform for indexing and searching text, vectors, tensors, and structured data, with machine-learning inference at serving time. Use it to build large-scale search, recommendation, personalization, and RAG systems.

RAG | Java · AI search · vector search

Code-Graph-RAG parses a multi-language monorepo with Tree-sitter, stores functions, classes and relationships in a Memgraph knowledge graph, and answers natural-language queries that generate Cypher. It also edits code via AST patches, finds dead code, merges runtime traces, and runs as an MCP server for Claude Code.

RAG | Python · knowledge-graph · code-analysis

OpenKB is a Python CLI that compiles raw documents (PDF, Office, HTML, CSV, URLs) into a persistent interlinked markdown wiki using LLMs and PageIndex vectorless tree retrieval. It offers query/chat with citations, a Skill Factory for agent skills, a web UI and REST API, all via LiteLLM providers.

RAG | Python · llm · knowledge-base

Production-grade Java Agentic RAG platform covering document ingestion, multi-channel retrieval (vector/keyword/graph/web) with RRF and rerank, intent recognition, query rewrite, session memory, MCP tool calling and a ReAct agent engine. Ships a React admin console, tracing and enterprise reliability features.

RAG | Java · agentic-rag · mcp

Hyper-Extract is a Python CLI and library that turns unstructured documents into structured knowledge — graphs, hypergraphs, temporal and spatial graphs — using LLM extraction engines like GraphRAG, LightRAG and Hyper-RAG, with FAISS-backed semantic search, incremental provenance tracking and an optional MCP server.

RAG | Python · knowledge-graph · hypergraph

nano-graphrag is a compact (~1100 LOC) MIT-licensed Python GraphRAG implementation: inserts text, builds an entity/relation graph, and answers global or local queries. Pluggable LLMs, embeddings, vector stores (FAISS, Milvus, Qdrant, hnswlib) and graph stores (networkx, Neo4j), fully async.

RAG | Python · graphrag · knowledge-graph

Python GraphRAG library that builds and incrementally updates domain-specific knowledge graphs, then uses PageRank-based exploration to retrieve evidence for queries. Supports OpenAI-compatible models, Gemini, and Vertex AI.

RAG | Python · GraphRAG · knowledge graphs

SimpleMem is a lifelong memory stack for LLM agents: it stores dialogues and multimodal inputs as compressed atomic memories with embeddings, then retrieves them semantically. Ships a Python package, MCP server, cross-session memory, and a self-evolving retrieval tuner. MIT, Python.

RAG | Python · long-term-memory · llm-agents

PipesHub is a self-hostable enterprise knowledge platform that connects 50+ business systems to AI. It offers permission-aware search with verified citations, GraphRAG retrieval, and exposes the same context to agents via MCP and Python/TS/Go SDKs.

RAG | Python · enterprise-search · permission-aware

Knowhere is an open-source document parsing and retrieval backend that converts PDFs, Office files, images, and text into hierarchy-native chunks with citations and cross-document links. It serves this memory to AI agents and RAG pipelines, including via MCP.

RAG | Python · agentic-rag · document-parsing

MySQL-compatible embedded/server database unifying vector, full-text and relational data in one engine. Two-level HNSW hybrid search, async index pipeline for streaming writes, and FORK/MERGE copy-on-write sandboxes make it a state store for agent memory and RAG.

RAG | C++ · vector-database · hybrid-search

Memobase is a user-profile-based long-term memory service for LLM apps: insert chat blobs, it batch-extracts structured profiles plus a time-aware event timeline, then returns prompt-ready context in <100ms. Self-hosted on FastAPI/Postgres/Redis with Python, Node, Go SDKs and an MCP server.

RAG | Python · long-term-memory · user-profile

Reference implementation of a controllable RAG agent where a deterministic LangGraph 'brain' plans, decomposes, retrieves and verifies to answer complex multi-hop questions over PDFs. Uses FAISS vector stores for chunks, chapter summaries and quotes, with Ragas evaluation and a Streamlit visualizer.

RAG | Jupyter Notebook · langgraph · langchain

Open-source Obsidian plugin adding hybrid semantic search, an auto-generated topic knowledge graph, and an agent that reads/writes notes with skills, memory and MCP. Search and graph work without an AI provider; embeddings unlock semantic mode.

RAG | TypeScript · obsidian · obsidian-plugin

An Azure solution accelerator for deploying secure agentic RAG applications. It combines Microsoft Agent Framework orchestration with Foundry IQ retrieval across enterprise data sources, with Azure AI Search as an alternative backend.

RAG | Python · agentic-rag · enterprise-rag

Java-based self-hosted enterprise search server on OpenSearch. Crawls web, file, DB and cloud sources with a browser admin UI, REST API, 20+ language analysis, and AI/RAG semantic search plus MCP support.

RAG | Java · enterprise-search · self-hosted

Collection of Jupyter notebooks and example apps showing AI-powered search with Elasticsearch as a vector database. Covers RAG, hybrid/semantic search, LangChain and OpenAI integrations, and document chunking for LLM applications.

RAG | Jupyter Notebook · vector-search · semantic-search

FastAPI + LangChain RAG service that embeds documents per file_id and stores them in PostgreSQL/pgvector (or Atlas MongoDB). It exposes async add/query/delete routes with JWT-verified owner scoping, built mainly as the retrieval backend for LibreChat.

RAG | Python · fastapi · langchain

FastAPI service that indexes documents with LangChain and stores embeddings in PostgreSQL/pgvector, exposing ID-based add/query/delete RAG endpoints. Built for LibreChat integration, with JWT-based per-user and entity ownership scoping on every read and delete.

RAG | Python · fastapi · pgvector

Self-hostable TypeScript engine plus Next.js app for AI-native document workflows: OCR ingestion, pgvector hybrid RAG, knowledge graph, LLM abstractions, and Inngest background jobs. Ships a startup-accelerator reference app with role-based access and missing-document detection.

RAG | TypeScript · pgvector · knowledge-graph

A PostgreSQL server and extension that automate embedding generation and upkeep, with APIs and SQL functions for semantic, full-text, and hybrid search. Use it to build RAG retrieval on PostgreSQL, including managed databases that cannot install extensions.

RAG | Rust · PostgreSQL · vector search

Memoir is a Python memory system for AI agents with hierarchical semantic paths, search, and Git-like branching, commits, and rollback. It includes a CLI, MCP server, and integrations for coding and assistant agents.

RAG | Python · agent memory · semantic search

TypeScript knowledge-base engine for the pi agent that ingests URLs, PDFs and markdown into an interlinked, Obsidian-compatible wiki, then serves it over MCP. Adds search, embeddings, linting and layered personal/project vaults so agent memory compounds instead of resetting each session.

RAG | TypeScript · knowledge-base · obsidian

MiniSearch is a self-hosted private search engine: SearXNG aggregates web results, a local cross-encoder reranks them, and an AI assistant writes cited answers using either in-browser LLMs (WebGPU/CPU via wllama) or an OpenAI-compatible backend. Ships as one Docker container, no API key or telemetry required.

RAG | TypeScript · web-search · in-browser-inference

Rust CLI that indexes past AI coding agent sessions (Claude Code, Codex, pi, Hermes) into a local Lance dataset, then serves hybrid vector+BM25 recall, MCP tools, and Hugging Face Hub publishing of the memory.

RAG | Rust · agent-memory · session-indexing

Python toolkit from AWS Labs for GraphRAG: builds hierarchical lexical graphs from unstructured data and composes graph-based question-answering strategies. Includes BYOKG-RAG for KGQA over your own knowledge graph, with integrations for Neptune, OpenSearch, PostgreSQL and LlamaIndex.

RAG | Python · graphrag · knowledge-graph

Curated awesome list of RAG research: surveys, papers, lectures, workshops and tools, with Semantic Scholar citation badges for each entry. Use it as a reading map for retrieval-augmented generation techniques and architectures.

RAG | awesome-list · retrieval-augmented-generation

RedisVL is the AI-native Python client for using Redis as a vector database. It covers index/schema management, vector, hybrid and filtered search, embedders, rerankers, semantic caching, LLM memory, semantic routing, plus a bundled MCP server.

RAG | Python · vector-search · redis

A local-first memory and reliability layer for AI agents, with hybrid retrieval, drift detection, and cross-agent coordination contracts. Provides MCP and A2A integrations for Claude Code and other agent clients.

RAG | Python · agent memory · long-term memory

Self-hosted AI backend that connects private files, databases, APIs, and MCP tools to local or hosted models through an OpenAI-compatible API. Includes RAG, model routing, authentication, guardrails, observability, and an admin UI.

RAG | Python · self-hosted · AI gateway

Open-source real-estate assistant combining conversational property search with ChromaDB-backed semantic and keyword retrieval. Includes listing analytics, valuation forecasts, neighborhood summaries, and a Next.js interface; supports multiple hosted LLM providers and Ollama.

RAG | Python · real-estate · property-search

LangChain integration package that implements core LangChain abstractions on Postgres via pgvector. Provides PGVectorStore (sync/async vector search, hybrid search, metadata filtering) and PostgresChatMessageHistory for persisting chat sessions.

RAG | Python · langchain · postgres

Memex is a local-first Rust CLI that indexes coding-agent transcripts (Claude Code, Codex, Cursor, OpenCode, Copilot, Pi) and searches them with BM25 plus optional embeddings for hybrid retrieval. It exposes the index through a TUI/web/native apps, an MCP server, and an installable agent skill, and can resume sessions,

RAG | Rust · claude-code · codex

Python RAG demo that indexes documents into Chroma, answers questions via LangChain models, and lets you visually explore question/snippet embeddings with Renumics Spotlight and UMAP to debug and evaluate retrieval quality.

RAG | Python · langchain · streamlit

Remnic provides local-first, Markdown-backed memory for AI agents, with hybrid search, provenance, correction and retrieval-quality controls. Agents access the shared store through MCP or HTTP integrations.

RAG | TypeScript · agent memory · long-term memory

An curated directory of blueprints, tools, research, and guides for building LLM-compiled knowledge bases. Covers agent-maintained Markdown wikis, persistent memory, graph-based approaches, and comparisons with traditional RAG.

RAG | LLM wiki · knowledge bases

Pure Rust PDF toolkit whose headline feature is structure-aware RAG chunking: each chunk carries pages, bounding boxes, element types, heading context and token estimates, with no ML or C dependencies. Also parses, generates, encrypts and validates PDFs in one crate.

RAG | Rust · pdf · chunking

YantrikDB is a Rust cognitive memory engine for AI agents: a vector/knowledge store that consolidates duplicates, detects contradictions, and decays stale memories. Ship it as an embeddable library, HTTP/HA cluster, or MCP server with 15 memory tools.

RAG | Rust · agent-memory · cognitive-memory

Docling Java is the official Java API for IBM's Docling document processing stack. It parses PDFs, Office files, images and audio into a unified DoclingDocument, exporting Markdown, HTML, DocTags or JSON for RAG and GenAI pipelines.

RAG | Java · docling · document-parsing

A .NET property graph database combining relational storage, HNSW vector search, and graph queries for AI knowledge retrieval. Includes grounded LLM chat and an MCP server so agents can work with graph data.

RAG | C# · graph database · vector search

CPU-only local memory layer for AI coding agents: tree-sitter plus local embeddings index code across repos, and a hybrid ranked kb_search returns trust-verified STRONG/WEAK/STALE hits. Integrates with Claude Code, Codex, Cursor, Windsurf and pi.

RAG | JavaScript · memory · cross-repo

Monorepo of official MongoDB + LangChain/LangGraph integration packages: Atlas vector, hybrid and full-text retrievers, semantic cache, chat history, plus LangGraph checkpointer and long-term memory store. Use it to wire MongoDB Atlas into RAG pipelines and agent memory in Python.

RAG | Python · mongodb · mongodb-atlas

Local-first RAG memory server for AI agents: hybrid BM25+vector plus GraphRAG search over docs and code, exposed via FastAPI REST API and an MCP server with OAuth 2.1. Ships a Claude Code plugin (30 commands, 3 agents, 2 skills) and installers for Codex, Cursor, Grok and OpenCode.

RAG | Python · agent-memory · graphrag

Local-first knowledge engine that indexes Markdown, PDF, Office files and code with hybrid BM25 + vector retrieval, cited LLM answers, and MCP/REST/CLI/SDK surfaces. Ships a Web UI, daemon, and one-command MCP install for ten AI clients; no GPU or cloud required.

RAG | TypeScript · local-first · hybrid-search

A local Rust daemon that stores and retrieves semantic memories for AI assistants, with a CLI, browser UI, and MCP interface. Uses LanceDB and Qwen3 embeddings, and integrates with Claude Code, Codex, Cursor, Zed, and OpenClaw.

RAG | Rust · semantic memory · local-first

Local-first hybrid search engine for Obsidian vaults, combining BM25, fuzzy title/alias matching and sqlite-vec semantic embeddings fused with RRF. Ships a CLI, an Obsidian plugin, and an MCP server so agents can search, read and traverse notes as tool calls.

RAG | TypeScript · obsidian · hybrid-search