Open-source repo-scale coding agent on AMD MI300X — 256K context, FP8. MIT. Reads your entire repo on a single GPU.
AI Engineer
Open-source repo-scale coding agent verified on AMD MI300X. Single GPU, Qwen3-Coder-Next-FP8, 256K context. 31/31 users at 8K-64K, 3/3 needle 200K, 9/9 correct on pytorch/vision (1.3M tokens). 124-min stress test, AITER tuning A/B. MIT.