kernel-design-agents
Kernel Design Agents (KDA) is an agent-centric workflow framework designed to guide coding agents through performance-critical CUDA kernel development. It provides structured prompts, task templates, and integration with specialized skills to help agents research, implement, verify, and iteratively refine GPU kernels. Developed by MIT's HAN Lab, KDA is an early research prototype that documents a systematic approach for AI agents to tackle low-level systems programming tasks requiring domain expertise and rigorous validation.
Key Features
Use Cases
- 01Developing optimized CUDA kernels for machine learning inference libraries
- 02Iterating on GPU kernel performance with agent-driven profiling and refinement
- 03Competing in MLSys kernel optimization contests with AI-assisted implementation
- 04Researching low-level systems programming approaches using coding agents
- 05Documenting and reproducing kernel optimization experiments with structured workflows
- 06Teaching agents to handle performance-sensitive code through structured task decomposition
Related Agents
View morehermes-agent
The agent that grows with you
agency-agents
A complete AI agency at your fingertips - From frontend wizards to Reddit community ninjas, from whimsy injectors to reality checkers. Each agent is a specialized expert with personality, processes, and proven deliverables.
openinterpreter
A coding agent for open models like Kimi K3
cline
Autonomous coding agent as an SDK, IDE extension, or CLI assistant.
kernel-design-agents — FAQ
What is Kernel Design Agents?+
Kernel Design Agents (KDA) is an agent-centric workflow framework that guides coding agents through the process of designing, implementing, verifying, and optimizing performance-critical CUDA kernels. It provides structured prompts, task templates, and integration points for domain-specific skills and profiling tools.
How do I install Kernel Design Agents?+
Clone the repository with submodules, then symlink the KernelWiki and ncu-report-skill into ~/.claude/skills/. You also need to install the humanize Claude Code plugin from the plugin marketplace. Finally, create a separate task workspace where the agent will perform the actual implementation work.
Which AI clients work with Kernel Design Agents?+
Kernel Design Agents is designed for Claude Code, as evidenced by the Claude Code plugin integration and the CLAUDE.md agent instructions file. The workflow leverages Claude's coding capabilities to iteratively develop and optimize CUDA kernels.
Do I need API keys or special hardware to use KDA?+
You need access to Claude Code (which requires an Anthropic subscription) and CUDA-capable hardware for kernel execution and profiling. The framework integrates with NVIDIA's NCU profiling tools, so you'll need the CUDA toolkit installed for performance analysis.
Is Kernel Design Agents free to use?+
The KDA framework itself is open source and free. However, you need a Claude Code subscription to run the agent workflow, and you'll need NVIDIA GPU hardware to execute and profile CUDA kernels.
What skills does KDA provide for the agent?+
KDA integrates two specialized skills: KernelWiki (GPU kernel optimization knowledge) and ncu-report-skill (NVIDIA NCU profiler report analysis). These skills give the agent domain expertise for CUDA development and performance debugging.
How do I install kernel-design-agents?+
Open the source repository on GitHub and follow its README. kernel-design-agents is a agent — MCP Agents Market links you directly to the official repo.
Is kernel-design-agents free?+
kernel-design-agents is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.