An AI assistant that answers from your documents, without letting your data out.
Your teams waste time looking for information.
What you get
- Answers that cite the exact source
- Installed on your premises, in France or elsewhere
- Word and PDF export
Try the demo
Ask a question about a fictitious company’s documents. Then click a source to see the exact passage.
Private assistant · demo
Suggested questions
On video
Decision tools
Two minutes, your own answers, a clear opinion.
Try it
Private LLM or API: the right choice for you
Your data is…
Do you have a GPU server?
You want to start…
Your tasks are mostly…
Our recommendation
Answer the four questions to see the recommendation.
Try it
GPU memory calculator
Weight precision
Text length per conversation
Simultaneous conversations
Estimated GPU memory
≈ 24 GB
- Model weights: 16 GB
- Conversations KV cache: 5.4 GB
- Runtime headroom: 2.6 GB
Example configuration: 1 × 48 GB GPU
Rough estimate: full KV cache for every conversation, in 16-bit, with no optimisation. Recent inference engines share memory across conversations: the real need is often lower. A token is a piece of a word.
Before / after
Before
- Manual searches through folders
- The same questions to the same colleagues
- Public assistants used on the quiet
- Sensitive documents sent outside
After
- A question in plain language
- Every answer cites its source
- An official tool, with accounts and roles
- Nothing leaves your infrastructure
How it works
- 01
Scoping
Documents in scope, users, available hardware and acceptance criteria.
- 02
Adaptation and evaluation
An open-source model (Mistral, LLaMA, Qwen or Gemma) adapted to your business, then evaluated on your cases.
- 03
Installation
On your premises or on our servers, in France or in your country, in Docker containers, with health checks and documentation.
- The result
- Your teams find information without digging, and your data stays with you.
- Proof before commitment
- A real demo of the platform, then a model evaluated on your cases before going into production.
- The timeline
- The pilot timeline is set and dated in the quote.
- Your effort
- We prepare the documents with you: sorting, formatting, anonymising if needed.
Entry offer
The Pilot: one repetitive task automated on your data, evaluated before any wider commitment.
- Scope, timeline and price set in the quote, in writing
- Acceptance criteria written down before we start
- If it fails the evaluation, it does not go into production
- One point of contact, who prepares the data with you
What is included
- Conversations with per-user history, answers streamed as they are written
- Analysis of PDF, Word, CSV and text, with sources displayed
- 16 pre-written prompt templates in 4 categories
- Export of an answer or a conversation to PDF and Word
- Optional web search, which can be disabled platform-wide
- Administration: accounts, roles, usage statistics
- Docker deployment, health checks, operations documentation
What is not
- No character recognition: a scanned PDF cannot be used
- No image, audio or video input
- No connection to the corporate directory (SSO)
- The model can make mistakes, even with sources: human review is still needed
Ideal for
- Companies whose information is spread across many documents
- Firms and engineering offices whose data rules out public assistants
Less suited for
- Archives that are only scanned, with no usable text
- A need to analyse images, audio or video
In 30 minutes, we look together at whether a pilot makes sense for you.