Based inParis

We build the AI systems
your company owns.

Generative AI engineering, RAG, agents, AI Act compliance, deployed on your own infrastructure.

CeresBrokerimmediatelyunderstoodourconstraintsandourobjectives.Theywereabletostructureatailoredsolution,reachingasystemperfectlyalignedwithourvisionandouroperationalneeds.

Jordan Béa
Jordan Béa
Studely

What we build.

Your documents become a knowledge base you can query in plain language. Hybrid search, answers cited down to the passage, connected to your tools through MCP: the assistant answers from your sources, never from memory.

Selected work.

Let's talk about your project.

Frequently asked questions

Assistants that answer from your document bases, agents that run business processes, automations, and the production layer that goes with them: evaluations, guardrails, oversight. We work with SMEs as well as regulated organisations: finance, healthcare, public sector, legal, industry.
A short scoping phase first: understanding your process, setting the architecture, signing an NDA if needed. Then sprints with a demo every week, directly on your infrastructure. Finally the handover: documentation, training for your team, code on your side. You can run the system without us.
Yes. On direct engagements, embedded with your technical team, or as a subcontractor for a consultancy or agency that owns the client relationship. Remote, or on site in the Paris region.
It belongs to you: the code, the data, the infrastructure, the documentation. Your team is trained to operate and evolve it. We remain available for maintenance and evolutions, but nothing obliges you: there is no technical dependency on our firm.
On European infrastructure (OVHcloud, Scaleway, Outscale), with SecNumCloud-qualified hosting or on-premise deployment for the most sensitive data. We deploy open models such as Mistral within your perimeter: your data never transits through a US cloud and remains out of reach of the Cloud Act.
They are designed to be: traceability, logging, technical documentation and data governance built in from the start, in alignment with the AI Act, GDPR and the ISO/IEC 42001 framework. The regulation continues to evolve, but our architectures are designed to remain compliant whichever deadline is set.
On your specific use case, yes, and often better. Production performance does not come from the base model but from the system around it: retrieval quality, evaluations, guardrails, orchestration. A well-designed system on an open model beats a simple call to a proprietary model, and you stay the owner of the whole thing.
Ceres Broker, sovereign generative AI engineering