Semantic Router
I worked on the core router for F-Secure and AMD Silo AI’s agentic AI demo, directing requests across local, private, and cloud models.
I worked on the core router for the agentic AI project demonstrated by F-Secure and AMD Silo AI. The project explores how to route requests across local, private, and cloud AI models while keeping consumers’ digital journeys safe and private.
Choosing where inference runs
The routing approach considers data sensitivity, cost, model capability, and performance to decide where a request should be processed and which model should handle it. A routine interaction can stay close to the user, while a more sensitive or demanding step can follow a different route.
The demonstration combines this routing with F-Secure’s trust assessment. Each step of a journey is evaluated in context: a trusted website can still contain untrusted reviews, hidden instructions, or sensitive checkout information. Lemonade provides the local AI serving layer in the architecture.
F-Secure plans to use the approach in TrustPath. The AMD announcement, published on 14 September 2026, describes the collaboration and its agentic shopping example.