apecloud/ApeRAG

ApeRAG: Production-ready GraphRAG with multi-modal indexing, AI agents, MCP support, and scalable K8s deployment

ApeRAG – Production‑ready Retrieval‑Augmented Generation platform

ApeRAG is an open‑source system that lets you build AI applications that can search and reason over your own documents. It combines several retrieval techniques (vector similarity, full‑text, graph‑based, summary‑based and vision‑based) with AI agents that can automatically pick the right source, run the query, and return a natural‑language answer. The project ships a FastAPI backend, a React web UI, and full Kubernetes/Helm deployment scripts, making it suitable for both local experimentation and enterprise‑grade production.

Core capabilities (as described in the README)

Feature	What it does
Hybrid index types	Provides five built‑in indexes – Vector, Full‑text, Graph, Summary, Vision – so the system can retrieve text, embeddings, graph relationships, concise summaries, or image content.
Intelligent AI agents	Agents are exposed via the Model Context Protocol (MCP) and can automatically discover relevant collections, run hybrid searches, and even perform web searches to answer questions.
Graph RAG with entity normalization	Uses a heavily modified LightRAG engine that merges duplicate entities, producing cleaner knowledge graphs and better relational reasoning.
Multimodal document processing	Handles PDFs, Word files, tables, formulas, and images; vision support lets the system analyse charts and pictures.
MinerU document parser	An optional parsing service (doc‑ray) that extracts complex structures (tables, formulas) with optional GPU acceleration.
Production‑grade deployment	Helm chart and KubeBlocks integration for PostgreSQL, Redis, Qdrant, Elasticsearch, and Neo4j; includes audit logging, model management, and graph visualisation.
MCP integration	Exposes a `/mcp/` endpoint so external LLM assistants can query the knowledge base directly using the Model Context Protocol.
Developer‑friendly stack	FastAPI backend, async Celery workers, React frontend, extensive docs, and a Makefile for quick Docker‑Compose starts.

Typical use cases

Enterprise knowledge bases – turn internal documents, manuals, and support tickets into a searchable AI assistant.
Product documentation portals – let users ask natural‑language questions and get answers drawn from PDFs, code docs, and diagrams.
Research libraries – combine text, tables, and figures; agents can retrieve relevant papers and summarise findings.
Customer‑support bots – hybrid search across FAQs, chat logs, and product images to provide accurate, context‑aware replies.

Getting started (local Docker‑Compose)

git clone https://github.com/apecloud/ApeRAG.git
cd ApeRAG
cp envs/env.template .env               # set your API key, DB passwords, etc.
# start all services (API, frontend, databases, optional doc‑ray)
docker-compose up -d --pull always

Web UI: http://localhost:3000/web/
OpenAPI docs: http://localhost:8000/docs
MCP client configuration example is provided in the README; just point the client to http://localhost:8000/mcp/ and supply the API key.

Production deployment (Kubernetes)

Prepare a K8s cluster (v1.20+), install kubectl and Helm 3.
Deploy databases – either use existing PostgreSQL/Redis/Qdrant/Elasticsearch instances or let the supplied KubeBlocks scripts spin them up.

Install the Helm chart:

helm install aperag ./deploy/aperag --namespace default --create-namespace

Expose services – port‑forward for testing or configure an Ingress for external traffic.
Optional doc‑ray service – enable in values.yaml (docray.enabled: true) or disable to save resources.

The chart handles all required containers (API, frontend, Celery workers, doc‑ray, etc.) and injects the correct connection strings.

Extending / contributing

The repo includes a development guide (docs/en-US/development-guide.md) that walks through setting up the codebase locally, running unit tests, and adding new agents.
The LightRAG component is open‑sourced under MIT; modifications are tracked in aperag/graph/changelog.md.
Community channels: Discord and Feishu QR codes are listed in the README.

License

Apache License 2.0 – permissive, allowing commercial use and modification.

Bottom line: ApeRAG is a full‑stack, production‑oriented RAG platform that blends multiple retrieval modalities with AI agents and offers both simple Docker‑Compose startup and scalable Kubernetes deployment. It is well‑documented, open‑source, and positioned for enterprises that need a self‑hosted knowledge‑base‑as‑an‑assistant.