GitHub weekly stars: DeepSeek open-sources DeepInfer inference engine
DeepSeek's newly open-sourced DeepInfer ranks second in weekly star gains; it runs V4.1 Flash on NVIDIA and Ascend chips but omits parts like the KV cache system.
The 5 hottest AI repos on GitHub this week. Ranked by stars gained Oct 19–25, via Trendshift. Non-AI projects skipped. 1. REA by morluto Reverse engineering for AI agents. Your agent calls REA through MCP, inspects an app or a native binary and gets back findings with evidence. Then it can explain how a feature works, or rebuild it and test it. 2. DeepInfer by DeepSeek The inference engine behind DeepSeek's own production services, now open source. It runs DeepSeek-V4.1-Flash on NVIDIA and Huawei Ascend chips. A full serving setup still needs parts this release leaves out, like the KV cache system. 3. security-audit by Cloudflare A skill that turns a coding agent into a security auditor. It maps the code, sends separate agents to hunt for bugs, then hands every finding to a fresh agent that tries to disprove it. Cloudflare says its internal vulnerability harness grew from this skill. 4. Harness² by Google Research Improves the system around a model instead of the model: prompts, tools, memory, skills. It proposes changes, tests them on the same task and keeps what works. The model weights never change. 5. AI Job Search A Claude Code workflow for job hunting. It scores job posts, tailors your CV, writes cover letters and preps you for interviews. The author built it after his position was cut and says it got him hired as an AI engineer. 4 of the 5 work on the layer around the model, not the model itself. That layer is where things move fastest right now. Links in the reply.