Skip to content

Design Partners

gpuemu's daemon, CLI, Python client, and VS Code extension are dual-licensed MIT / Apache-2.0 with no telemetry. They are and will remain so. The research backing (P1–P4) and the kernel corpus are public.

We are working with design partners on an enterprise tier to extend the OSS core with the surfaces a procurement department needs:

  • Private rule packs — your kernel families, your invariants, your tolerance envelopes, versioned and signed.
  • Signed Kernel Correctness Reports (HTML + PDF) suitable for SLA evidence packages, with ed25519 signatures customers can verify offline. See the inference-vendor walkthrough.
  • On-prem daemon with SSO (SAML / OIDC), audit logs, and air-gapped operation.
  • Custom CI integrations beyond the public GitHub Action — GitLab, Jenkins, Buildkite, Bazel, internal CI servers.
  • Support SLAs and a private Slack / Teams channel with the maintainers.

We are deliberately not publishing tier names or prices in advance. Pricing depends on the shape of the first piloted use case; the competitive research suggests five-to-six-figure annual contracts for frontier-lab kernel teams and inference-as-a- service vendors.

Who we want to hear from

  • Inference-as-a-service vendors that ship custom Triton / CUDA kernels and want to publish a signed correctness artefact with their SLA.
  • Frontier-lab kernel teams that want a pre-merge correctness gate scaled to 100s of ops with a fair-use site licence.
  • Maintainers of large OSS inference projects (vLLM, SGLang, TensorRT-LLM, llama.cpp, …) who want a sponsored OSS tier — corporate users paying for the project-specific rule pack and the badge.

Contact

Email: me@dipankar.name

A one-line description of what your kernels look like (CUDA / Triton / Pallas, rough op count, target GPU classes) is plenty for a first conversation. We do not collect or store usage telemetry; the only information we have is what you tell us.