antirez spent 15 years making Redis fast. Now he’s aimed that same obsession at local AI: ds4 (DwarfStar), a local inference engine written in C that runs DeepSeek V4 Flash/PRO and GLM 5.2 on your own hardware. 20.3k GitHub stars, +385 today, top of GitHub Trending.
Deliberately narrow, aggressively optimized
ds4 is not another general-purpose GGUF runner — it targets these few models and goes deep. Asymmetric 2-bit quantization on the MoE experts only. SSD streaming so a 64GB machine can run models bigger than its RAM. Distributed inference across machines (two M5 Max Macs get a 1.66x prefill speedup). A built-in coding agent and native tool calling, straight from the CLI. Backends for Metal, CUDA and ROCm. On a MacBook Pro M5 Max at Q2, it hits 463 tokens/s prefill and 26 tokens/s generation.
Why it matters
This is the Redis playbook — small, native, zero dependencies — applied to the messiest layer of local AI. And the README admits it: built “with strong assistance from GPT 5.5, 5.6, Claude Fable,” humans leading design and debugging. The creator of Redis, co-writing C kernels with AI, so you can run Chinese open models at home. That’s the whole 2026 stack in one repo.
You Might Also Like
- Title ds4 Deepseek v4 Metal Local Inference Engine by Antirez Redis Creator Runs v4 Flash on a Single Macbook
- 27k Github Stars in Weeks Learn Claude Code by Shareai lab Breaks Down ai Coding Agents Into 12 Lessons
- From Claude Flow to Ruflo 22k Stars 5900 Commits and the Multi Agent Swarm Taking Over Claude Code
- Deepseek tui Tops Github Trending a Claude Code Clone Wired to Deepseeks api
- Zcode Zhipu z ai glm 5 2 Coding Agent hit hns 1 Spot Chinas Open Answer to Claude Code
Top AI Product 
Leave a comment