Priced around what
you actually ship.
Kernalin is in early access and pricing is not finalized. Both plans start with a conversation — tell us your GPU workloads and we'll scope the right fit.
Early Access
Early access openFor teams evaluating Kernalin for production GPU optimization.
- Workload intake + static security checks
- Verification engine — correctness gate, anti-cheat, statistical verdict
- Benchmark regression gate + GitHub Action / CI
- Profiling sessions + run history
- Memory bottleneck analysis (7 dimensions)
- AI optimization copilot
- Triton, PyTorch, and ONNX workloads; CUDA C++ in preview
Enterprise
For organizations running GPUs at scale, with security and compliance requirements.
- Everything in Early Access
- Enterprise security review Roadmap
- SSO (SAML 2.0 / OIDC)
- RBAC + workspaces
- Custom / self-hosted infrastructure Roadmap
- Observability dashboards + alerts Preview
- Dedicated support Roadmap
Pricing is not finalized. Figures and plan boundaries may change while Kernalin is in early access.
Pricing, plainly.
Is pricing final?
No. Pricing is not finalized. Both plans are quoted through a conversation while Kernalin is in early access — there are no published prices yet.
What's available today vs roadmap?
Available: sign-up and workspaces, workload intake, the verification engine, benchmarking, profiling, memory analysis, the AI copilot, the CI regression gate with GitHub integration, alerts, RBAC and SSO (OIDC and SAML). Preview: cost monitoring, live GPU dashboards, PyTorch → Triton conversion, CUDA C++ optimization and the VS Code extension. Roadmap: enterprise security review, self-hosted deployment and dedicated support.
Which frameworks are supported?
Triton and PyTorch are fully supported; ONNX models can be uploaded and benchmarked; CUDA C++ optimization is in preview. You bring the kernel or workload; the engine handles intake, verification, and benchmarking.
How does the CI gate work?
A GitHub Action runs the benchmark regression gate on each PR and blocks a merge if a change is convincingly slower, with an auditable metric-diff attached to the run.
Not sure which plan fits?
Tell us about your GPU workloads and we'll scope the right early-access engagement.