TokenSpeed is an LLM inference engine built for agentic workloads, with a C++ scheduler control plane and Python execution plane.
Java framework for building distributed, long-running agents with an event stream, tool-call permissions, middleware, and sandboxed workspaces.
Enterprise low-code platform (Spring Boot and Vue3) that generates frontend, backend, and database code from natural-language prompts or a code generator.