Inco Splash: open-source Apple silicon inference engine, 282ms first token on M5 Pro
aigclink · x · 2026-09-20
The GitHub project incoai/splash is now open source: a local inference engine for Apple silicon, built around the model, serving a small set of models to coding agents and any OpenAI/Anthropic-compatible client.
- On a 48GB M5 Pro it decodes Qwen3.8-27B at 2x the next-fastest engine, returning the first token in 282ms with a 32K cached context
- Its speed comes from per-model specialized kernels, a draft model, and a memory plan — hence nothing to configure
- Requirements: Apple M3+, macOS 26.4+, 36GB unified memory (48GB+ recommended)
- Quick start: brew install incoai/tap/splash, then splash serve --model incoai/Qwen3.8-27B-Splash; it verifies the model and serves on 127.0.0.1:8000
More from coding & agent
- Veteran game dev: "AI builds a full game in a day" posts are mostly hype — bendee983 · 2026-11-04
- Dev builds interactive Excel-to-dashboard with ChatGPT, shares the prompt — TermRemarkable665 · 2026-11-04
- Agents cut dev time from weeks to a day; bottleneck moved to specs and review — Suspicious_Orchid770 · 2026-11-04
- Orgo, a YC-backed platform giving AI agents their own cloud desktops, claims 1,000+ teams — Scobleizer · 2026-11-04
- Founder: Vibe-coded a PMF-successful product, then hired pro devs to fix the mess — sujingshen · 2026-11-04
- One prompt to audit cross-region traffic: dev cut trans-Atlantic latency by moving a service — DanielLockyer · 2026-11-04