We build AI infrastructure for teams who do not want to send their tokens to someone else's cloud.
Every "AI gateway" we looked at in 2025 wanted us to send our prompts and outputs through their data plane. That's a bad shape for compliance-bound workloads, sensitive customer data, and any team running their own GPUs for cost reasons. So we built the inverse: the gateway lives next to your model, our cloud only orchestrates.
Hux is short for hook — the agent hooks every request before the GPU sees it, the way a fishing hook intercepts what swims by. Kan is கண் (kaṇ), Tamil for eye. The Kan control plane is the eye that watches over every Hux you've enrolled. Together: hooks on the GPUs, one eye in the cloud watching them all.
Bangalore, India. Building globally.