by kekzl · Agent Tool · ★ 29
From-scratch CUDA inference engine for the NVIDIA RTX 5090 (sm120a). The best single-GPU backend for agentic AI — tool calling, long-context loops, reasoning and concurrent sub-agents, on top of the fastest single-user inference on the 5090: faster decode than llama.cpp (+37–72% dense GGUF), at-or-ahead of vLLM on NVFP4, and the only engine running native NVFP4 on consumer Blackwell. 97k lines, 100% written by Claude Code. What is imp A C++20/CUDA inference engine that targets exactly one architecture: the NVIDIA RTX 5090 / RTX PRO 6000 (GB202, ).
| Stars | 29 |
| Forks | 2 |
| Language | Cuda |
| Category | Agent Tool |
| License | MIT |
| Quality Score | 62.0996217167065/100 |
| Last Updated | 2026-07-08 |
| Created | 2026-02-23 |
| Platforms | claude-code |
| Est. Tokens | ~19k |
These tools work well together with imp for enhanced workflows:
Explore other popular agent tool tools:
imp is From-scratch C++/CUDA inference engine for the NVIDIA RTX 5090 (sm_120a) — the best single-GPU backend for agentic AI: tool calling, long-context loops, reasoning and concurrent sub-agents on top of t. It is categorized as a Agent Tool with 29 GitHub stars.
imp is primarily written in Cuda. It covers topics such as blackwell, cpp, cuda.
You can find installation instructions and usage details in the imp GitHub repository at github.com/kekzl/imp. The project has 29 stars and 2 forks, indicating an active community.
imp is released under the MIT license, making it free to use and modify according to the license terms.