News/Open Source
Try nowOpen Source·IncrementalOfficialbreakingUpdated Oct 2·Updated 1×·Event Oct 5, 2026·First seen Oct 5

Redis Creator Salvatore Sanfilippo Releases ds4 Local LLM Inference Engine

LaunchOpen sourcePractical
Try it

Something you can actually use or run today.

On October 5, 2026, Salvatore Sanfilippo released ds4, a high-performance C-based local inference engine supporting Metal, CUDA, and ROCm.

It provides a minimalist, low-dependency alternative to llama.cpp for developers who need extreme control over memory and hardware kernels.

AILookup take

When the creator of Redis builds a systems tool, you pay attention. This won't replace mainstream tools for most, but for engineers building specialized edge devices, this is a goldmine of optimized C code.

Who cares
systems engineersembedded AI developerslocal LLM enthusiasts
Watch next

Integration of ds4 kernels into other popular inference wrappers like Ollama.

Also covers

  • Redis Creator Launches ds4 for Local LLM ExecutionSalvatore Sanfilippo, the creator of Redis, has released a new tool called 'ds4' designed for running Large Language Models locally.
Details
  • Provides a lightweight, low-dependency alternative to frameworks like llama.cpp for C-based applications.
  • Offers optimized inference performance specifically tailored to DeepSeek V4.1, Qwen3.8, and GLM 5.x architectures.
  • Serves as a high-quality reference for systems-level implementation of KV-cache management and hardware kernels.
Consensus

Multiple sources confirm the release of the ds4 engine by Salvatore Sanfilippo and its focus on specific LLM architectures.

DwarfStar 4 (ds4): Local DeepSeek V4.1, Qwen and GLMHacker News· 1 stories
AILookup

Research utility for AI tools. Compare sourced profiles, distinguish listings from automated research, and track tool changes without marketing fluff.

© 2026 AILookup. All rights reserved.