Sovereign enterprise AI chat
A fully sovereign enterprise chat assistant: self-hosted model, data that never leaves the infrastructure, fine-grained access control. Project in its early phase.
Tech stack
- Mistral (self-hosted)
- vLLM
- RAG
- pgvector
- RBAC
- On-premise / France
- FastAPI
Sovereignty requirements
Define hosting, governance and access constraints: no data leaves the company’s infrastructure.
Target architecture
Self-hosted open-weight model, a RAG layer over internal sources, role-based access control (RBAC) and logging.
Pilot
A small initial scope to validate answer quality, performance and adoption before a wider rollout.
Screenshots
Screenshots coming soon — project under NDA.