Projects

We turn persistent agent systems research into public systems, reusable tooling, and deployment-ready infrastructure.

They close the loop between frontier ideas, engineering execution, and real-world application feedback.

Flagship Project

vLLM Semantic Router

vLLM Semantic Router is an open-source routing system designed to send each request to the right model and execution path across local, private, and frontier deployments.

Unified routing across heterogeneous model pools and deployment environments.
Decision surfaces informed by cost, latency, privacy, safety, and workload semantics.
A bridge from inference-time routing into broader agentic system orchestration.

Project website GitHub

Type Open-source routing system

Scope Inference, model selection, and agent-facing control flows

Focus Cost, latency, privacy, safety, and execution quality

How projects work.

Projects are the lab's way of turning technical questions into systems that can be measured, used, and maintained in public.

Research Fit

Start from a concrete systems question.

Each project should capture a real problem in agentic intelligence, not just package a prototype.

Build Surface

Ship a reviewable public surface.

Projects need maintainers, interfaces, benchmarks, and clear implementation boundaries.

Deployment Loop

Learn from real workloads.

Projects improve by closing the loop between research hypotheses, engineering constraints, and use.