openTPU: building an open-source AI accelerator on FPGA with Claude Code agents
ScarionnS · reddit · 2026-10-15
A Redditor open-sourced openTPU, an AI accelerator developed end-to-end with coding agents (mostly Claude Code), already running Qwen3.5 and other open-weight models on a Kintex-7 FPGA.
Multi-agent hardware optimization loop:
- One agent reads HDL, timing reports, and lessons from prior attempts, then proposes a change
- Another implements it
- Automated checks diff hardware against a Python simulator, test performance, and simulate results
Lessons: giving agents precise feedback (which test failed, where the timing bottleneck is, whether it helped) and keeping that loop cheap is key; big architectural changes still need human nudging; verification has blind spots — floating-point bugs slipped past LLM-inference tests.
The repo includes hardware code, an instruction set, simulator, a compiler for Python-style kernels, a cycle-level profiler, and host tools. No FPGA needed — run models through the simulator on a laptop. Open source: github.com/FeSens/openTPU
More from coding & agent
- Claude cut a 2.6-hour video export to 20 minutes — but only after being told to rebuild the server — byhewar · 2026-10-23
- 3 flight-booking rules my AI agent checks that I'd never do by hand — Efistoffeles · 2026-10-23
- OpenJet claims Qwen3.8-27B coding runs 1.9x faster than llama.cpp on one RTX 3090 — JeremyCMorgan · 2026-10-23
- Building an autonomous SEO agent in Hermes with three skills: programmatic pages, Keywords Everywhere, DataForSEO — EXM7777 · 2026-10-23
- Building a sports betting AI agent with onchain execution and human approval by default: architecture and lessons — Dry-Comfortable-2678 · 2026-10-23
- How do you audit AI agent decisions in regulated workflows? A field survey — Ok_Way_1993 · 2026-10-23