SentinelleIA is built by Apex Numeris
Apex Numeris is a French deeptech company based in Lyon. We build runtime security and governance for production AI agents: software that sits in the traffic, authorizes each action, and produces evidence a regulator or an auditor can read.
Why a network team builds this
SentinelleIA is an inline component on production traffic. Interception, authorization before execution, fail-closed behaviour, latency budget, reversibility by disabling a route. Those are network architecture problems before they are AI problems.
The founding team are telecom network architects, not AI researchers. Two decades of running critical platform migrations at a European tier-1 operator, where a control that fails open is an incident and a millisecond is a budget line. That is the discipline we apply to agent traffic: MCP and A2A are one more protocol plane to intercept, authorize and log.
What we build
SentinelleIA
Runtime security and governance for agentic AI. One authorization boundary for every agent action, interception of MCP and A2A traffic, and a SHA-256 chained audit log aligned with Articles 9 to 15 of the EU AI Act.
See the platform →Sovereign by construction
Container based deployment on your own infrastructure. The supervisor is separate from the agents it governs and keeps its own audit chain. No outbound dependency, no call home, no data leaving your perimeter for a decision to be made.
See the frameworks →Where we are
The platform runs in production on sovereign infrastructure, at TRL 6. We publish what we measure, including the cost in false positives, and we state plainly what is industrialised and what is still being industrialised.
We are open to first European design partners, in regulated sectors and among critical operators. The first level of our assessment requires no access to your systems and is free.
Talk to the team
Questions on the platform, on evidence for the EU AI Act, or on a design partnership.
Contact us