# Nativ：在 Mac 上本地运行前沿开放模型 - 来源：Hacker News 热门（buzzing.cc 中文翻译） - 作者：aratahikaru5 - 发布时间：2026-07-21 05:55 - AIHOT 分数：67 - AIHOT 链接：https://aihot.virxact.com/items/cmrtsfndu22tfbihzuhc83gk9 - 原文链接：https://blaizzy.github.io/nativ ## AI 摘要 Nativ 发布 macOS 桌面应用，让用户在 Apple Silicon（M1+）上本地运行 Google、Cohere、Liquid AI 等开放模型，无需账户或订阅。应用支持语言、视觉、视频、代码、音频等多模态交互，并提供实时 token/秒、内存压力等性能指标。Nativ 完全开源（MIT 许可），内置本地端点，可对接 Claude Code、Codex 等编码智能体。 ## 正文 Run AI models locally on your Mac. Nativ puts frontier intelligence on your desk. Download and run open models on Apple Silicon — no accounts, no subscriptions, no cloud. Download for macOS Join the Discord Universal · Apple Silicon (M1+) Nativ, running on macOS Real app. Real local models. No cloud. 01 / CHAT Talk to open models with streaming responses and per-message performance metrics. CAPTURED LIVE ON THIS MAC · JUL 2026 / SIX REASONS TO GO LOCAL Everything you need. Nothing you don’t. 01 CURATED LIBRARY Pick the right model for your Mac Run standout open models from Google, Cohere, and Liquid AI. Nativ recommends the right partner model for your hardware. 02 CHAT Prompt it like you would Claude A clean interface with streaming, markdown, code highlighting, and image input. Every response is generated locally. 03 TELEMETRY See what’s actually happening Live tokens/sec, memory pressure, thermal state, and time-to-first-token. The details developers want. 04 APPLE SILICON Optimized for Apple Silicon Built on MLX-VLM and tuned for M-series unified memory and Metal — no wrappers, no translation layers. 05 EVERY MODALITY Language, vision, video, code, audio Chat with an LLM, caption an image, summarize video, autocomplete code, or transcribe and generate speech. 06 FREE FOREVER Not a SaaS. Not a subscription. No accounts to create, no credits to buy, no data to sell. You own it end-to-end. MIT LICENSED / INTEGRATIONS Your tools. Your models. Connect the coding agents you already use to models running locally on your Mac. Nativ handles the endpoint; your workflow stays familiar. 01 Pi 02 Codex 03 Claude Code 04 Hermes 05 OpenCode LOCAL ENDPOINT One model server. Every coding agent. MANIFESTO · REV 1.0 Why we’re open source when nobody else is. philosophy.txt UTF-8 $ cat philosophy.txt The other “local AI” apps you’ve heard of? They’re proprietary shells built on top of open-source engines they don’t own. They keep the UI closed, add a paywall, and hope you don’t look under the hood. We built in the open. The desktop app is open too. Every line. Every model loader. Every telemetry chart. You can read it, fork it, or send a pull request tonight. No VC roadmap. No enterprise tier. No dark pattern that turns your prompts into training data. Just software made by researchers and hackers, for researchers and hackers. / PARTNER MODELS Open models from teams we trust. Google Cohere Liquid AI MODEL CREATOR CONTEXT SIZE TYPE Gemma 4 E2B Instruct Google 128k 10.28 GB VISION + AUDIO ↗ North Mini Code Cohere 500k 19.38 GB CODE + TOOLS ↗ LFM2.5-VL 1.6B Liquid AI 128k 3.20 GB VISION + LANGUAGE ↗ Explore the full model library → YOUR MAC IS MORE CAPABLE THAN YOU THINK Stop renting intelligence. Run it locally, in the open. Download for macOS ↘ View source ↗