AI inference, rebuilt from first principles
Infervisor builds the software that serves, routes and meters AI models on GPUs you own.
What we build
- plow, the inference engine. It compiles each model ahead of time for the GPU it will run on. Open source under Apache-2.0.
- InferOS, the control plane. It runs a fleet of GPU machines as one inference platform. In private beta.
- Plow SMI, GPU monitoring and power control for NVIDIA, AMD, Intel and Apple. Open source.
- Model registry, signed, prebuilt models that install with one command. Free.
How we work
- The engine is open source, so anyone can read what runs their models.
- Our published comparisons link to settings and source reports and include results where vLLM is ahead. See benchmarks.
- Accuracy comes before speed: results are checked against a reference before any timing counts.
Team
- Lava Kumar Bokam Founder LinkedIn profile
Get in touch
Talk to an engineer about your models and GPUs, or follow our work.