IntelliDocs
An AI-native, fully self-hosted enterprise document management system — upload, search, and ask.
An AI-powered enterprise Document Management System that goes well beyond storage. Users upload, organize, and search documents — and ask natural-language questions answered directly from their content through retrieval-augmented generation. Everything, including LLM inference, runs self-hosted on the organisation’s own GPU server.
Enterprises need to search and reason over sensitive documents without shipping them to a third-party AI API. Keyword search alone isn't enough — but a private alternative has to run retrieval, embeddings, and LLM inference entirely in-house, absorb heavy indexing work without blocking users, and stay locked down behind enterprise-grade security.
AI-native document Q&A (RAG)
Instead of keyword search alone, users ask natural-language questions and get answers grounded in the actual content of their documents, powered by a retrieval-augmented generation engine.
Live streaming responses
AI answers stream to the UI over a WebSocket connection, giving the responsive, chat-style “typing” experience users expect rather than a long wait for a full reply.
Full-text + semantic search
Elasticsearch delivers fast keyword and full-text search across every document, complementing the AI-driven semantic retrieval so nothing gets lost.
Scalable background processing
Celery workers handle heavy jobs — document processing, indexing, embedding generation, OCR — asynchronously, so the user-facing app never blocks on them.
Private & self-hosted by design
Documents, embeddings, and even LLM inference stay on the organisation’s own GPU server (an RTX 4000 Ada) via Ollama — nothing leaves the network, which is critical for sensitive enterprise content.
React + Vite frontend
An “Enterprise DMS” single-page app served as static assets, talking to the backend over a REST API and a WebSocket for streamed AI responses.
FastAPI backend
A Python FastAPI application server (via uvicorn) exposes the REST API and a WebSocket endpoint (/ws/) for live-streamed AI output.
RAG & LLM layer
RAGFlow powers document-grounded Q&A, LiteLLM provides a unified proxy across LLM providers/models, and Ollama runs local models on the GPU.
Polyglot data layer
PostgreSQL holds core app data, Elasticsearch indexes documents for search, MinIO (S3-compatible) stores the files themselves, Redis handles caching and the Celery broker, with MySQL backing the RAGFlow component.
Layered security & infrastructure
nginx handles reverse-proxy and TLS termination in front of every service, ModSecurity (OWASP CRS) adds a web-application-firewall layer, and the GoDaddy SSL certificate holds an A rating on SSL Labs.
Modular, containerised services
Each function — search, storage, LLM, queue — runs as its own container, so any component can be scaled or swapped independently without touching the rest.
Have a project like this?
Tell us what you're building — thirty minutes with an engineer, not a salesperson.