Your AI document assistant, hosted in France
The brain that answers based on your documents powered by MaxKB, installed and maintained by DINAO. An AI knowledge base (RAG) where your documents stay on our French servers — in local mode, they never go to OpenAI.
What is MaxKB?
MaxKB is an open-source platform for enterprise-grade intelligent agents, combining a knowledge base and a question-answering assistant based on large language models (LLM) and RAG (Retrieval-Augmented Generation). It enables the creation of chatbots and agents capable of querying a document base to reduce hallucinations.
The RAG pipeline imports documents directly or crawls web pages, then automatically segments and vectorizes the text into a PostgreSQL database with the pgvector extension. MaxKB is model-agnostic: it connects to local models (DeepSeek, Llama, Qwen via Ollama) as well as external APIs (OpenAI, Claude, Gemini), and orchestrates agent workflows with tool calls via MCP (Model Context Protocol).
Multimodal (text, image, audio, video), MaxKB integrates « zero code » into third-party systems via an embeddable Q&A widget, and exposes an API. Built on Python/Django and Vue.js with the LangChain framework, under GPLv3 license, it deploys with a single Docker command.
Host MaxKB at DINAO
Resource tiers compatible with MaxKB prerequisites (minimum 2 vCPU / 2048 Mo / 5 Go). Hosted in France, fully managed.
- 2 dedicated vCPU
- 4 Go RAM
- 40 GB NVMe
- Daily backups
- Managed & monitored by DINAO
- 4 dedicated vCPU
- 8 Go RAM
- 80 GB NVMe
- Daily backups
- Managed & monitored by DINAO
- 8 dedicated vCPU
- 16 Go RAM
- 160 GB NVMe
- Daily backups
- Managed & monitored by DINAO
1 hidden tier(s) (insufficient resources for this app) : Découverte
This application uses AI
The container hosts the application, not the AI engine (which requires dedicated GPUs) : inference is handled externally, with your own provider key. Prioritize a sovereign engine — Mistral AI, NumSpot, Scaleway, or OVHcloud AI Endpoints (France, GDPR) ; an international provider (OpenAI, Anthropic…) only if a specific capability requires it. Inference subscriptions are not included in hosting.
Technical details
You might be wondering…
Are my documents used to train a model?
No. Your documents and questions remain private on your instance. In local mode (Ollama), no data is transmitted to a third party or used for training.
Which AI models can I use?
MaxKB is model-agnostic. For sovereignty, we prefer a local open-weight model (DeepSeek, Llama, Qwen…) served by Ollama. It is also possible to connect an external API (OpenAI, Claude, Gemini), but data then leaves the French perimeter.
How does the AI answer based on my documents?
MaxKB uses a RAG pipeline: your documents (or crawled web pages) are vectorized in a PostgreSQL/pgvector database. The AI relies on this content to answer, which significantly reduces hallucinations.
Can I integrate it into my site?
Yes. MaxKB offers an embeddable « zero code » Q&A widget, as well as an API and MCP tools for more advanced integrations.
Is a GPU required?
Not necessarily. A small model or external API works without a GPU. For better performance with a larger local model, a GPU (optional or dedicated) is recommended — DINAO advises you based on your volume.
