Perplexity open-sourced their Mac inference server for Qwen 3.6

Products·building

Bobcat — local LLM inference engine optimized for Apple Silicon Macs

Bobcat is a local LLM inference engine optimized across the stack for consumer MacBooks on Apple Silicon. It runs models locally and claims to be the fastest way to run models like LFM, Qwen3.5, and Clef Flash on Mac. It targets Mac users who want fast local inference without cloud services.

Bobcat is a local LLM inference engine optimized across the stack for consumer MacBooks on Apple Silicon. It runs models locally and claims to be the fastest way to run models like LFM, Qwen3.5, and Clef Flash on Mac. It targets Mac users who want fast local inference without cloud services.

Releases·Tool Release

Qwen3.8-27B at 144 tok/s on an M5 Max MacBook Pro

Inco released Splash, an open-source inference engine built for Apple silicon, running Qwen3.8-27B at up to 144 tok/s on an M5 Max MacBook Pro. It claims up to 3× Ollama's decode speed and 2× oMLX, with nearly 4× gains when agents fan out into sub-agents. It ships as a CLI server plus an LM Studio runtime, targeting local agent work.

Inco released Splash, an open-source inference engine built for Apple silicon, running Qwen3.8-27B at up to 144 tok/s on an M5 Max MacBook Pro. It claims up to 3× Ollama's decode speed and 2× oMLX, with nearly 4× gains when agents fan out into sub-agents. It ships as a CLI server plus an LM Studio runtime, targeting local agent work.

A developer released qMLX, a custom inference engine for Qwen 3.5 122B on Apple Silicon, extending MLX with hybrid attention support, SSD-backed KV cache, and RYS layer duplication. It enables efficient serving of the hybrid MoE model on consumer Mac hardware.

Releases·Tool Release

Draw Things Releases Public Beta of Local Code, a Local Inference Engine for Coding Agents

The Draw Things team opened a public beta of Local Code, a coding agent built on the Draw Things inference engine and distributed via TestFlight. It claims 1.2x-1.6x faster prefill than alternatives on Apple Silicon, with no sign-up required. The app ships as a sandboxed, self-contained package bundling common coding tools.

The Draw Things team opened a public beta of Local Code, a coding agent built on the Draw Things inference engine and distributed via TestFlight. It claims 1.2x-1.6x faster prefill than alternatives on Apple Silicon, with no sign-up required. The app ships as a sandboxed, self-contained package bundling common coding tools.

A developer released Samosa Chat, a tool that enables running the Qwen3.6-35B-A3B model locally on a 16 GB Mac. The project is hosted on GitHub and was shared on Hacker News on July 15, 2026.