AI Potluck
Contribute
Back to Gap Map Model components / Inference code

ds4

antirez (Salvatore Sanfilippo)
open source / Overall score: 3.0

ds4 (DwarfStar 4) is a native local LLM inference engine by antirez (Salvatore Sanfilippo), optimized first for DeepSeek V4 Flash and deliberately narrow rather than a general GGUF runner. It also runs DeepSeek V4.1 Flash, GLM 5.x and Qwen3.8 Flash Next. The C/CUDA/Objective-C engine draws on llama.cpp/GGML kernels and quantization formats, adding model loading, prompt rendering, tool calling, and KV state streaming to RAM and SSD. It ships an HTTP server with OpenAI-style and Anthropic-style endpoints and a coding agent.

The project has no tagged releases and is built from source, so the entry is read against the repository head rather than a release.

Openness

5 high confidence
5.0
license
MIT(OSI)
source
public
core features withheld
no

MIT-licensed native inference engine; full C/CUDA/Obj-C source public, no proprietary tier.

  • https://github.com/antirez/ds4 recorded 2026-08-09

    repo license metadata reports {"spdxId":"MIT","name":"MIT License"}; single public repository - a full-text read of the page finds no commercial edition, enterprise directory or hosted-service offering, the only "pricing" strings being GitHub's own site navigation, and the project links no separate product site.

  • https://github.com/antirez/ds4/blob/main/LICENSE recorded 2026-09-16

    MIT License, Copyright (c) 2026 The ds4.c authors, Copyright (c) 2023-2026 The ggml authors

  • https://raw.githubusercontent.com/antirez/ds4/main/LICENSE recorded 2026-09-29

    The repository LICENSE file is the MIT License text, opening with "MIT License" and "Copyright (c) 2026 The ds4.c authors" and "Copyright (c) 2023-2026 The ggml authors".

  • https://raw.githubusercontent.com/antirez/ds4/main/README.md recorded 2026-09-29

    The README describes the full engine source in the repository, saying "The code is self-contained and deliberately narrow, not a general GGUF runner" and "Some source-level pieces are retained or adapted here under the MIT license"; a search for enterprise, commercial, license key and hosted found nothing gating the engine.

Adoption

3 low confidence
3.0

GitHub stars are the only usage signal published for ds4, since no download, install or customer figure exists. A star is not a use, so this measure can only be read as a rough floor on adoption.

Capability

3 medium confidence
3.0

ds4 is a narrow engine that runs a short list of chosen open-weight models rather than a general GGUF runner. It has neither the model breadth of llama.cpp nor the datacenter throughput of frontier serving engines.

  • https://github.com/antirez/ds4 recorded 2026-08-09

    README states the engine is "deliberately narrow, not a general GGUF runner. Model loading, prompt rendering, tool calls, KV state, the HTTP server, and the coding agent are built and tested together"; confirms Metal/CUDA/ROCm backends and DeepSeek 4 Flash/PRO focus

  • https://raw.githubusercontent.com/antirez/ds4/main/docs/SERVER.md recorded 2026-09-29

    The serving guide lists "POST /v1/chat/completions" as "OpenAI-style chat" and "POST /v1/messages" as "Anthropic-style messages".

  • https://raw.githubusercontent.com/antirez/ds4/main/README.md recorded 2026-09-29

    The README says the engine is "deliberately narrow, not a general GGUF runner" and that "model loading, prompt rendering, tool calls, KV state, the HTTP server, and the coding agent are built and tested together"; it lists Metal, CUDA and ROCm targets and DeepSeek V4, GLM and Qwen model support.

Verified 2026-09-27