HAgent an open framework to build hardware agents
summary
HAgent is an open-source framework (https://github.com/masc-ucsc/hagent) that integrates LLMs with chip design tools using a compiler-like pipeline of hermetic passes that exchange state via YAML interfaces. It runs LLM blocks (Gemini/OpenAI) alongside an MCP server to do tasks like patching generated Verilog from Chisel, with caching, incremental steps and formal checkers; current results show roughly 80% correct patches for the Chisel→Verilog repair flow, while token cost and split of work between agent and MCP remain open research problems.
AI-written summary · about
chapters (8)
- 0:00HAgent framework intro
- 1:25Motivation for hardware agents
- 3:09LLMs for hardware design
- 4:35Verifiable agent pipelines
- 6:00Incremental steps and caching
- 7:03Demo: compiling hardware designs
- 9:53Conclusions and future work
- 10:15Q&A
related talks
40mAgents in Action: Advancing Open Source AI Missions
40mBuilding Agents with Agent Bricks and MCP
01h 43mBuilding LLM Agents (Hackathon)
01h 42mBuilding Responsible AI Agents in Decentralized Open-Source Environments
36mPutting AI Agents to Work: A New Era of Open Source Connectivity with MCP
29mFrom Copy-Paste to Composition: Building Agents Like Real Software
01h 03mAI Coding Agents and how to code them
23mGoverning AI Agents at the Hardware Boundary