Hoid AI

Hoid provides a software stack for AI inference, designed to help teams run models faster and more efficiently across different hardware. The company describes its product as an agentic AI compiler: a bridge between a model and its target silicon that profiles performance, identifies bottlenecks, benchmarks alternatives, and searches for optimizations. Hoid combines a low-latency runtime for specific hardware providers with an agentic harness that continuously tests and applies improvements, rather than relying solely on fixed compiler heuristics or manual, chip-specific tuning.

The platform is intended for inference providers and teams seeking stronger performance from AI accelerators, including those with existing in-house inference systems. Hoid supports both open-source and custom models. For a pilot, customers select a model and target hardware, then agree with Hoid on performance goals relevant to their workload. The company says its approach can help shorten optimization timelines as new models emerge and make better use of available hardware.