Local vs Cloud

You stay in control

Decide at any time whether your data stays with you or benefits from the power of the secure European cloud. The search engine and the interface are identical in both modes.

🔒 100% Local mode

Indexing and LLM on your machine. No data sent over the Internet.
  • Absolute confidentiality: no data sent out
  • No external server for processing
  • Local LLM via llama.cpp (GGUF Mistral, Llama…)
  • Local indexing (embeddings + BM25 + reranking)
  • Ideal for sensitive data (industry, legal, medical)
  • Modest PC: slower indexing and queries
  • Recommended: 16 GB RAM, NVIDIA GPU for good performance
Included in every licence

☁️ 100% Cloud mode

Online LLM, indexing on our European servers. Best performance.
  • Online LLM (Mistral API): maximum quality
  • Offloaded indexing on OVHcloud (France)
  • No hardware constraints on the user side
  • Strongly accelerated indexing and queries
  • Data encrypted in transit (TLS), hosted in the EU
  • Ideal for large volumes, light setups, distributed teams
Requires a higher plan

How it works

Three steps, that's all

From your PDF folder to your research assistant, in a few minutes.

1

Point to your library

Your Zotero or Mendeley folder, or any folder of PDFs. Nothing to upload.

2

Smart indexing

GROBID parsing, OCR, semantic chunking, embeddings, BM25, topics — in the background.

3

Chat with your articles

Natural-language questions, sourced answers with clickable citations.

Decision aid

How to choose?

A few practical rules based on your profile and constraints.

You are…Recommended mode
A researcher with a recent PC + NVIDIA GPULocal — good performance, zero recurring cost beyond the licence
A firm handling confidential files (legal, medical)Local — mandatory, no data should transit
An industrial R&D team (patents, internal reports)Local or Team (hybrid) depending on sensitivity
A PhD student with a modest PC, 5,000+ articlesCloud — otherwise local indexing can take days
A library of 30,000+ articlesCloud — local becomes uncomfortable beyond ~20,000
A distributed team sharing an indexCloud or Team — centralised index on OVHcloud FR

The right reflex

Start local on a sample (a few hundred PDFs) to validate the use case. Switch to cloud or hybrid mode (Team plan) if your volumes or hardware constraints call for it — RefChat handles the migration without manually re-indexing everything.

Not sure which mode fits?

A 30-minute chat is enough to point you the right way — no commitment.

Contact us