Kog

Kog develops the Kog Inference Engine (KIE), a hardware-software stack for running large language model inference on GPUs. The site describes it as a drop-in replacement for vLLM and includes the Kog Communication Library (KCCL), a replacement for standard GPU communication layers used in tensor parallelism.

Customers are presented as developers of AI coding agents and agentic workflows that need faster generation for multi-step reasoning. Access is offered through a "Request API Access" form, so the business model appears to be an API service; the site does not publish pricing.

The site states the company is based in Paris, France, and that its team includes PhDs, GPU and research engineers, with founders from X, Centrale Supélec and ENS. It lists backing from Varsity, CNRS and BPIFrance and received the French Tech 2030 label in October 2025. Open roles listed are Agentic Compiler Engineer and Research Engineer - LLM Architecture, both in Paris.

Product: Kog Inference Engine (KIE), a drop-in vLLM replacement for faster LLM inference, with the Kog Communication Library (KCCL) for multi-GPU tensor parallelism.