🔒 100% Local available

🇫🇷 Made in France

🛡️ GDPR-friendly

Chat with your scientific library

RefChat is a research assistant for your scientific PDFs. Hybrid search, automatic query mode detection, multilingual, built-in OCR. Your choice: 100% local or secure EU cloud.

Request a license See modes

Why RefChat?

A research assistant built for scientific rigor and data sovereignty.

🔒

Absolute privacy

Your documents never leave your machine in local mode. Cloud mode available if you'd rather offload compute to our European servers.

🎯

Precise hybrid search

Multilingual semantic search (E5-large), BM25 keyword and cross-encoder reranking. More relevant than basic RAG, with explicit source citations.

🧠

5 smart modes

RefChat automatically detects what you want: question, summary, references list, author search, fact-checking.

📄

Advanced OCR & parsing

Scanned PDFs? Old image reports? EasyOCR + GROBID extract text, metadata and scientific structure.

🗂️

Automatic topics

BERTopic clustering organizes your library into coherent themes. Rename, merge, split as you see fit.

🌍

Native multilingual

Ask a question in English about a French paper (or vice-versa) — RefChat retrieves relevant content regardless of language.

Two usage modes, your choice

You stay in control. Decide at any time whether your data stays on your machine or benefits from cloud performance.

🔒 100% Local Mode

Indexing and LLM on your machine. No data ever sent over the internet.

Absolute privacy: no data sent to the internet
No external server used for processing
Local LLM via llama.cpp (GGUF models: Mistral, Llama, etc.)
Local indexing (embeddings + BM25 + reranking)
Ideal for sensitive data (industrial, legal, medical)
Lower-end PC: indexing and queries will be slower
Recommended: 16 GB RAM minimum, NVIDIA GPU for best performance

Included in all licenses

☁️ 100% Cloud Mode

Online LLM, indexing on our European servers. Maximum performance.

Online LLM (Mistral API): top-tier answer quality
Indexing offloaded to our OVHcloud servers (France)
No hardware constraints on the user's side
Drastically faster indexing and query response
Data encrypted in transit (TLS) and hosted in the EU
Ideal for large volumes, light setups, distributed teams

Requires a premium plan

Three steps, that's it

From your PDF folder to your personal research assistant in minutes.

Point to your library

Select your Zotero folder, Mendeley, or any folder containing your PDFs.

Smart indexing

Multi-threaded pipeline: GROBID parsing, OCR, semantic chunking, embeddings, BM25, topic clustering.

Chat with your papers

Ask questions in natural language. Sourced answers with clickable citations.

5 query modes auto-detected

Mode	Trigger	Behavior
Question	(default)	Narrative answer with sourced citations
Summary	"summarize", "synthesis"	Summary of one or more papers
References	"which papers", "list references"	List of relevant papers with explanations
Author	"papers by", "work of"	Author-based search via OpenAlex
Fact-check	"verify", "is this correct"	Statement verification against your library

Who is it for?

Built for demanding environments where privacy and rigor are paramount.

🔬 R&D and academic research

Literature reviews, cross-referencing hundreds of papers. Accelerate state-of-the-art reviews.

📊 Technology watch

Industry reports, patents, publications. Ask strategic questions without data leakage risk.

📚 Industrial knowledge base

Find information buried in decades of internal reports (OCR + RAG).

🏛️ Consulting & expertise firms

Ideal for organizations handling sensitive documents (legal, medical, geoscience).