Pulsar

Pulsar is an inference engine for running AI models on Apple Silicon.

Pulsar is an inference engine designed to run Qwen3.8-27B and Qwen3.6-35B-A3B AI models on Apple Silicon Macs. It is a fork of Inco AI's Splash 1.3.0 and adds new GPU kernels, drafting, and prompt lookup capabilities. The engine pairs a full model with a draft model to propose tokens, checking every guess to maintain accuracy while increasing speed. It supports chat in a browser, coding agents, and speaks both OpenAI and Anthropic APIs. It runs locally on macOS with specific memory and storage requirements.

Pulsar at a glance

Best for: Software engineer
Runs on: macOS
Pricing: Free
Released: October 2026
Category: AI coding tool
Website: github.com/loopai-hq/pulsar

Key features

Runs Qwen models locally on Apple Silicon
Pairs full models with draft models for speed
Provides prompt caching for fast repeated prompts
Speaks OpenAI and Anthropic APIs
Supports browser-based chat and coding agents

Use it for

Running local AI models for coding
Chatting with AI models in a browser
Connecting coding agents to local models

Alternatives to Pulsar

Kapitsa — Optimized inference engine delivering high token generation speeds on Apple Silicon.
oMLX — It uses paged SSD KV caching to reduce agent TTFT to under 5 seconds.
Strata — Runs large AI models locally on consumer gaming PCs without server reliance.
Lumabri — Runs large AI models across multiple computers by sharing memory and layer workloads.
fal — Access 1,000+ generative models and high-performance serverless GPUs.
Shimmy — Runs local AI models with zero Python or C++ dependencies.
AirLLM — Runs massive 70B+ parameter models on small 4GB consumer GPUs.
MTPLX — It runs local models twice as fast via native MTP speculative decoding.
Narwhal — It automatically hot-swaps prefill and decode roles without reloading model weights.
TokenSpeed — Optimized performance for agentic workloads and high-throughput LLM serving.
Union.ai — Runs machine learning tasks natively on Apple Silicon GPUs using Metal.
Rapid-MLX — Run 250 MLX-native models locally on Apple Silicon with high inference speeds.

All Pulsar alternatives