SourceBento
EN
Support Browse products
Business apps · AI + OCR

DocBento

Self-hosted document management that reads every scan and answers with page citations.

AI document management with 4 OCR engines, hybrid search and a chat assistant that cites the exact page. PostgreSQL or MariaDB.

Platforms
  • Web
Built with
  • Python 3.12
  • FastAPI
  • SQLAlchemy
  • React 19
  • TypeScript
  • Tailwind
  • PostgreSQL + pgvector
  • MariaDB 11.8
  • Docker
DocBento preview
Highlights

What DocBento does

Four OCR engines

PaddleOCR PP-OCRv6 (offline on CPU), Tesseract 5, Docling for layout and tables, and Vision AI.

Answers with citations

An agent-based assistant with 13 skills that answers questions and links to the exact page.

Hybrid search

Full-text plus vector search across languages, filtered by what each user may see.

Fields filled for you

AI summary, document type, tags, expiry dates and custom key fields per document type.

Word templates

Templates with {{placeholders}}, filled by hand or by AI.

Your choice of database

PostgreSQL with pgvector or MariaDB 11.8 with native vectors, chosen at install.

DocBento is a self-hosted document management system with AI built in. Drop in PDFs, scans, photos and Office files. DocBento extracts the text, runs OCR where needed, indexes everything for keyword and semantic search, and fills in the summary, document type, tags, dates and key fields. Your team can then ask questions and get answers that cite the exact page.

Built for real document work

  • Side-by-side view of each scan and its OCR text.
  • Reviews and approvals, expiry reminders, comments, versions and trash.
  • Roles, groups and restricted folders that also apply to search and the assistant.
  • Audit log, processing queue and a REST API with personal tokens.
  • Interface in English, Spanish, French and German.

An assistant that shows its sources

The assistant works like an agent: it finds documents by metadata, reads the relevant pages, searches passages and then answers. Its 13 skills include summarize, compare, extract, obligations, compliance checks, proofreading, number analysis with charts, drafting and translation.

What you receive

The complete source code (FastAPI + SQLAlchemy backend, worker, React 19 + TypeScript frontend), Docker Compose with an HTTPS overlay, backups, sample data, a self-resetting demo mode and documentation.

Requirements: Docker 24+, 2 CPU / 4 GB RAM minimum (4 CPU / 8 GB recommended). An AI API key is optional.

Screenshots

See it in action

DocBento: Assistant answer with page citations
Assistant answer with page citations
DocBento: Scan and OCR text side by side
Scan and OCR text side by side
DocBento: Hybrid search
Hybrid search
DocBento: Numbers analysed with charts
Numbers analysed with charts
DocBento: Word templates
Word templates
DocBento: OCR engine settings
OCR engine settings
Try before you buy

Open the DocBento demo

Sample data only. Nothing you do there affects anyone else.

Open the live demoSample documents with real AI answers. Resets every hour.
FAQ

Questions about DocBento

Anything else? Email [email protected].

Does it work without AI?

Yes. Upload, OCR (PaddleOCR, Tesseract, Docling), search, templates, reviews and permissions all work offline. AI is optional and works with any OpenAI-compatible provider, including a local Qwen-VL on Ollama.

Which languages does the interface support?

English, Spanish, French and German.

What server do I need?

Docker 24 or newer, 2 CPU and 4 GB RAM at minimum; 4 CPU and 8 GB RAM are recommended.

Do restricted folders also apply to search and the assistant?

Yes. Roles, groups and restricted folders are enforced in search results and in what the assistant can read.

News

Related reading

All news
More from SourceBento

You might also need

All products
Soon

BandBridge

Cross-platform desktop workbench for wearable development.