# DwarfStar 4

> MIT in-process coding agent — project quants only, shared KV cache

**Category:** AI Agents  
**Pricing:** open-source  
**URL:** https://github.com/antirez/ds4  
**GitHub:** https://github.com/antirez/ds4  
**Added:** 2026-10-03

---

## The Hook

DwarfStar 4 is the antirez in-process coding agent that only runs its own quants — DeepSeek V4/V4.1, GLM 5.x, Qwen3.8 — and shares the KV cache with the agent instead of shelling out to a generic local runner.

## What's Good

MIT (antirez/ds4). Raw LICENSE is the API. Free path is git clone, make, and download_model.sh against public Hugging Face GGUFs — no subscription. The coding agent runs in-process and shares KV cache with the model. Beta, with a healthy star pace (★~22,980 in about 172 weeks, ~134 a week).

## The Catch

Not a generic GGUF runner. Project quants only: DeepSeek V4/V4.1, GLM 5.x, Qwen3.8. Vs Ollama / Magnitude / llmfit: Ollama is the general local model runner, Magnitude is the consumer profiler plus agent-harness wiring, llmfit only sizes models to hardware — DwarfStar 4 is the antirez in-process coding agent with a shared KV cache, not a general local runner. Still beta; pin a commit before a fleet bet.

## Verdict

MIT in-process coding agent when you want the antirez shared-KV quants — not Ollama, not Magnitude, not llmfit.