🚀 DocuMind

Private AI for Your Documents (Nextcloud + RAG)

Ask questions to your private files — get accurate answers with sources, fully self-hosted.

---

✨ Why DocuMind?

Most AI tools require uploading sensitive data to external APIs.

DocuMind keeps everything local.

• 🔐 100% private (self-hosted)
• 🧠 AI-powered search across documents
• 📎 Answers with source citations
• 🏢 Built for teams, companies, and power users

---

🎯 Who is this for?

• 🧑‍💻 Developers building RAG / AI systems
• 🏢 Companies with sensitive documents
• 🔐 Privacy-focused teams avoiding SaaS AI
• ☁️ Nextcloud users wanting AI search

---

⚡ Demo (What you get)

Q: "What are the payment terms in our vendor contracts?"

→ Answer:
"Net 30 payment terms are defined in Section 4..."

Sources:
- /contracts/vendor_a.pdf (page 3)
- /contracts/vendor_b.pdf (page 2)

---

🧠 Core Features

• 🔎 Semantic search (RAG)
• 📄 Multi-format ingestion (PDF, DOCX, MD, TXT)
• 🔐 Permission-aware retrieval (Nextcloud ACLs)
• 🔄 Auto sync + reindex pipelines
• 🤖 Local LLM support (Ollama)
• 📊 Document intelligence (entities, risks, deadlines)
• 🧩 Nextcloud-native integration

Built on a scalable ingestion + classification pipeline

---

🏗️ Architecture (Simple)

flowchart LR
    NC[Nextcloud] --> API[FastAPI]
    FE[React UI] --> API
    API --> DB[(Postgres + pgvector)]
    API --> Redis
    API --> Worker[Celery]
    Worker --> AI[Embedding + LLM]

Loading

---

🚀 Quick Start (1 command)

NextCloud bridge: Copy nc_ai_bridge folder to Nextcloud extra apps folder(/var/snap/nextcloud/current/nextcloud/extra-apps/) to access app via icon on Nextcloud:

Docker:

make docker-dev

Local:

make local-dev

Then open:

• 🌐 App → http://localhost:5173
• 📡 API → http://localhost:8000/docs
• ☁️ Nextcloud → http://localhost:8081

Login:

Backend credentials (set in backend/.env):

FIRST_SUPERUSER_EMAIL=admin@admin.com
FIRST_SUPERUSER_PASSWORD=12345678

---

🔄 How It Works

Nextcloud → Sync → Parse → Chunk → Embed → Store → Query

Under the hood:

• ingestion pipeline
• classification engine
• RAG retrieval + citation builder

---

🔌 Integrations

• Nextcloud (WebDAV + OCS)
• Ollama (local LLMs)
• PostgreSQL + pgvector

---

🧪 Developer Friendly

make docker-dev
make docker-logs
make local-dev
make local-backend-test

Clean modular architecture:

• ingestion pipeline
• classification system
• AI layer isolation
• worker queues

---

🔐 Privacy First

• No external AI APIs required
• Runs fully local
• Respects document permissions
• Secure credential handling

---

⚠️ Limitations

• No OCR (yet)
• Limited file types
• Requires local model setup for best results

---

⭐ Why Star This Project?

• 🔥 Build your own private ChatGPT for documents
• 🧩 Plug into your existing Nextcloud instantly
• 🏗️ Production-ready architecture (not a toy demo)
• 🧠 Extendable AI pipelines (classification, workflows, agents)

---

🛣️ Roadmap

• OCR + image parsing
• Email ingestion (IMAP)
• Knowledge graph
• Multi-tenant SaaS mode
• Agent workflows

---

🤝 Contributing

PRs welcome. Open an issue first for major changes.

---

📄 License

MIT License — free for personal and commercial use.

---

💬 Final Note

DocuMind is not just another RAG demo.

It’s a foundation for building private AI knowledge systems on top of your existing infrastructure.