whereismyai
Hyperkernel
Hyperkernel is an open-source software implementation running compiled Megakernels for AI models on NVIDIA GPUs.
This software contains public implementations of Megakernels generated by Hyperkernel, currently in Alpha. It requires NVIDIA B300 GPUs and host RAM, running on a command-line interface. Supported models and tasks include DeepSeek-V4.1-Flash for LLMs, openpilot for driving, Evo 2 7B for DNA generation, Matrix-Game 3.0 for world models, ACT and pi0-FAST for robotics, and ESMFold for protein folding. Users must provide model checkpoints as safetensors directories. It compiles kernels for specific tasks and hardware configurations, offering OpenAI-compatible server endpoints and specialized processing pipelines.
Hyperkernel at a glance
Key features
- Run OpenAI-compatible server on 4 GPUs
- Execute openpilot driving model over camera frames
- Generate DNA sequences with Evo 2 7B
- Generate video clips using Matrix-Game 3.0
- Run ACT and pi0-FAST for robotics control
- Fold proteins using ESMFold
Use it for
- Serving large language models
- Processing driving model camera frames
- Generating DNA from prompt sequences
- Running robotics control ticks
- Performing protein folding tasks
Alternatives to Hyperkernel
- Strata — Runs large AI models locally on consumer gaming PCs without server reliance.
- OpenCode — Works with 75+ model providers, local models, and existing subscriptions
- fal — Access 1,000+ generative models and high-performance serverless GPUs.
- HyperQwen — High-throughput Qwen model serving on single consumer GPUs.
- GPUStack — Unifies AI model serving and GPU compute management across heterogeneous hardware.
- ForgeKernel — It continually self-evolves GPU kernels past the best known implementations.
- RunThisLLM — Provides exact hardware and VRAM requirements for over 295 local AI models.
- Daytona — It provisions sandboxes in under 90 milliseconds with full security isolation.
- OpenSandbox — Provides isolated environments with multi-language SDKs and lifecycle management.
- GPUWorks — Manages NVIDIA, AMD, Ascend, PPU, Hygon, and Apple Silicon in one cluster.
- Tunr — Deploys agent-built software and shares local dev servers instantly via a single command.
- Pebrel — It brings local shells, remote hosts, and AI command-line tools into one workspace.
All Hyperkernel alternatives