Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM
Easiest Desktop Apps

Nativ Review: Free Open-Source Local AI App for Mac with MLX

·7 min readBy · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

Nativ is a free, open-source (MIT) Mac app that runs open AI models on Apple Silicon through a bundled MLX server and adds chat, image generation, dictation and a local API for coding tools. It needs a Mac with an M1 or later chip and macOS 26 or newer, installs from a DMG or Homebrew, and, according to the maker, works offline once a model is downloaded. It is a young project from a Polish company. The maker describes it as free and account-free, although the terms mention that paid plans such as Nativ Pro might come later. PromptQuorum has not tested it.

Key Takeaways

  • Open source under MIT and free; the maker says there is no account and no subscription, and the terms say paid plans such as Nativ Pro would show their terms before any purchase.
  • Runs only on Macs with Apple Silicon (M1 or later) and macOS 26 or newer; MLX does not run on Intel Macs.
  • Features: chat with image input, image generation and editing, speech models, dictation in any app, meeting notes, scheduled prompts and MCP servers with per-tool consent.
  • A local server on port 8080 offers OpenAI- and Anthropic-compatible endpoints, and the app sets up coding agents such as Claude Code, Codex, OpenCode, Aider and Goose to use it.
  • PromptQuorum has not tested the app. When PromptQuorum checked on 18 November 2026, the privacy policy contained a bracketed payment-processor placeholder and sections on accounts, licence keys and purchases, although the maker describes the app as free and account-free.

📍 In One Sentence

Nativ is a free, open-source Mac app that runs open AI models on Apple Silicon and adds chat, images, dictation and a local API for coding tools.

💬 In Plain Terms

Think of it as a ChatGPT-style app that runs on your own Mac instead of a company's servers. You pick a model, it downloads once, and after that chatting, dictating and generating images work without an internet connection, according to the maker.

📌Note: The privacy policy says Neywa Labs does not collect your chats, prompts, files or outputs, and that diagnostic data is collected only with the consent you give at first launch. The app still goes online to download models from Hugging Face and to check for updates.

Nativ is a free, MIT-licensed Mac app from Neywa Labs that runs open AI models on Apple Silicon through a bundled MLX server. It offers chat with image input, image generation, dictation and meeting notes, and a local OpenAI- and Anthropic-compatible API that coding agents can use. This review is based on the maker's website, its privacy policy and terms, the GitHub repository and its release page, checked on 18 November 2026. PromptQuorum has not tested the app hands-on.

Nativ Review: Free Open-Source Local AI App for Mac with MLX

Overview

Nativ is a SwiftUI Mac app that wraps a bundled mlx-vlm server and the MLX runtime, so open models run on Apple Silicon without a separate server to set up.

The project is MIT-licensed and developed in the open at github.com/Neywa-Labs/nativ by Neywa Labs, a company the terms of use name as Neywa Labs sp. z o.o. in Kraków, Poland. The maker describes itself as the team behind the open-source libraries mlx-vlm and mlx-audio. The repository was created in July 2026. The latest release, v0.4.6, appeared on 16 November 2026 and the last commit was on 17 November 2026, so development is active and no stale-project warning applies.

According to the README, the app combines a chat window, a model library, an Images tab, an audio extension for dictation and meeting notes, a developer page with the local API, analytics, a system monitor and menu-bar controls. It also lets you open documents, code and web pages in a work pane beside the chat.

These descriptions are maker statements. PromptQuorum has not run the app and has not checked speed, answer quality, image quality or dictation accuracy.

How to Get It

You install Nativ like any Mac app: download the DMG and drag it to Applications, or use Homebrew.

The GitHub release v0.4.6 ships a single Nativ-0.4.6.dmg. The README also gives a Homebrew command, and the Homebrew cask list shows version 0.4.6. After installation the app updates itself through Sparkle, according to the README.

Route
How to get it
DMGDownload, drag Nativ to Applications
WebsiteDownload for macOS button
Homebrewbrew install --cask nativ
Sourcemake xcode-generate, make xcode-run (Xcode needed)

On first launch the README lists optional steps: pick or download a model, set a server API key, and grant microphone, accessibility and screen-recording permissions now or later. PromptQuorum did not run the installer or scan the download.

Requirements

Nativ needs a Mac with Apple Silicon (M1 or later), macOS 26 or newer and enough unified memory for the model you choose.

The list comes from the README and the FAQ, which says MLX does not run on Intel Macs. No app-level memory floor is published, and the model library shows fit warnings for your Mac, according to the README. PromptQuorum did not measure memory use or speed.

Item
Stated figure
ChipApple Silicon (M1 or later)
Operating systemmacOS 26 or newer
MemoryEnough unified memory for the model
InternetOnly for model downloads and updates
Hugging Face tokenOptional, for gated models
Building from sourceXcode, xcodegen, Python 3

The README says model downloads and first-time build dependencies still need network access. Optional permissions (microphone, accessibility, screen recording) switch on dictation, meeting capture and screenshots, and the FAQ says you can skip any of them.

Features

Nativ's core is a local chat app and model manager, with images, audio, a local API and agent integrations around it. All items below are maker claims from the README and FAQ.

  • Chat and vision: streaming conversations, image attachments, reasoning output, response metrics and persistent chat history, with several independent windows sharing one server.
  • Images: a dedicated tab that generates and edits images with compatible MLX image models.
  • Model library: browse and download compatible Hugging Face models with memory-fit warnings, move the cache to an external APFS drive, and preload language, image, speech and embedding models together.
  • Local API: OpenAI-compatible chat, responses, image, audio, embeddings and model endpoints plus Anthropic Messages endpoints on 127.0.0.1:8080 by default, with an optional API key for management routes.
  • Coding-tool integrations: set-up for Codex, Claude Code, Pi, Hermes, OpenCode, Aider, Goose, Crush, Qwen Code and OpenClaw, and for VS Code, Cursor, Zed, JetBrains, Cline and Continue.
  • Audio extension: dictation in any app, transcript history, meeting notes with a transcript and summary, and export to PDF, DOCX or TXT; raw dictation audio is deleted after five minutes.
  • MCP servers with per-tool consent, scheduled prompts with task kits, an artifacts library with semantic search, a system monitor and advanced inference controls such as KV-cache quantization and speculative decoding.

Privacy and Data Flow

According to the maker, your chats, prompts, files and outputs stay on your Mac; the app goes online to download models and check for updates, and diagnostic data is opt-in.

According to the privacy policy (last updated 16 November 2026), Neywa Labs does not collect, access or store your content, does not use it to train or evaluate models, and does not sell personal information. Diagnostic data such as crash reports, app and macOS version, chip and memory size, performance figures and feature-usage events may be collected only with the consent the app requests on first launch, and you can withdraw it in Settings. The policy says it never includes the content of your sessions or which specific models you use.

The FAQ adds that models are downloaded straight from Hugging Face and that update checks reach Neywa Labs or its update provider, which may see request details such as an IP address and app version. The website uses PostHog analytics with a cookie banner.

The policy also contains a payment-processor placeholder in square brackets and sections on registration, licence keys and purchases, although the maker describes the app as free and account-free. The terms add that paid plans such as Nativ Pro might come later. Read both documents again before relying on them, and note that the code is MIT-licensed and public, so the data flow can in principle be inspected; PromptQuorum has not done so.

Who Should Use It

Nativ fits Apple Silicon owners who want a free, open-source local AI app without an account.

  • Mac users who want chat, image generation and dictation in one app that runs models on their own machine.
  • Developers who want Claude Code, Codex, Aider or an editor to use a local model through OpenAI- or Anthropic-compatible endpoints.
  • People who want to browse and download MLX models from Hugging Face with memory-fit warnings.
  • Readers who prefer a project whose app code and licence are public.

What We Could Not Verify

Nativ is a poor fit in some situations, and several details remain unverified.

  • Not a fit: Windows, Linux or Intel Mac users. The app runs only on Apple Silicon with macOS 26 or newer.
  • Not a fit: people who need a long track record. The repository was created in July 2026 and the version is still 0.4.x.
  • Unverified: speed, memory use, answer quality, image quality and dictation accuracy. PromptQuorum has not run the app.
  • Unverified: that nothing leaves the Mac beyond model downloads and update checks. The claim is the maker's, and PromptQuorum did not audit the code or inspect network traffic.
  • Unverified: how the privacy policy's account, licence and purchase sections and the "[PAYMENT PROCESSOR NAME]" placeholder fit a free app, and what "Nativ Pro" would include.
  • Unverified: the maker's statement that it is the team behind mlx-vlm and mlx-audio, the number of models that run in the app, and user numbers. No independent reviews were found.
  • Unverified: whether the DMG is notarized and what minimum memory a given model needs on your Mac.

Nativ vs Alternatives

Nativ competes with other local desktop apps and Mac inference tools; PromptQuorum has not compared it with them hands-on. The tools below are in the directory and are described from the reviews on this site.

Compare in directory

FAQ

Is Nativ free?

Yes, per the maker. The code is MIT-licensed, and the FAQ says there is no account and no subscription. The terms say that if paid plans such as Nativ Pro are offered, their price, billing and cancellation terms will be shown before you buy.

Which Macs does Nativ run on?

Macs with Apple Silicon (M1 or later) running macOS 26 or newer. The FAQ explains that Nativ uses the MLX framework, which does not run on Intel Macs, so Intel Macs, Windows and Linux are not supported. PromptQuorum has not tested it on any Mac.

Can I use Nativ with Claude Code or Codex?

Yes, per the maker. Nativ serves the loaded model through OpenAI- and Anthropic-compatible endpoints, and its Integrations page sets up Claude Code, Codex, Pi, Hermes, OpenCode, Aider, Goose, Crush, Qwen Code and OpenClaw, plus editors such as VS Code, Cursor, Zed and JetBrains. PromptQuorum has not tested the set-up.

How do I install Nativ?

Download the DMG from the website or the GitHub releases page and drag Nativ to Applications, or run brew install --cask nativ. The app then updates itself through Sparkle. PromptQuorum did not run the installation.

Does Nativ send my chats anywhere?

Not by default, per the maker: chats, prompts, files and answers stay on your Mac. The app goes online to download models from Hugging Face and to check for updates, and diagnostic data is sent only if you consent at first launch. When PromptQuorum checked on 18 November 2026, the privacy policy also contained a bracketed placeholder, so read it before relying on it.

What can Nativ do besides chat?

According to the maker, it has an Images tab that generates and edits images with compatible MLX image models, an audio extension for dictation in any app and meeting notes with a transcript and summary, and support for speech models. PromptQuorum has not tested any of them.

Where do Nativ models come from?

From Hugging Face, through the in-app model library, which shows memory-fit warnings for your Mac, according to the README. A Hugging Face token is optional and only needed for gated models. The terms say each model has its own licence and you are responsible for following it.

Does Nativ work offline?

According to the maker, chatting, dictating and generating images work without an internet connection once a model is downloaded. Model downloads, first-time build dependencies and update checks still need the network. PromptQuorum has not tested offline use.

What does the Nativ local API offer?

According to the README, OpenAI-compatible chat, responses, image, audio, embeddings and model endpoints plus Anthropic Messages endpoints, served on 127.0.0.1:8080 by default, with an optional API key for management routes. PromptQuorum has not tested the API.

Does Nativ support MCP servers?

Yes, per the maker: you can connect MCP servers and expose their tools to chat, with per-tool consent. The README also lists scheduled prompts with task kits. PromptQuorum has not tested the MCP support.

Who makes Nativ, and how mature is it?

The terms of use name the provider as Neywa Labs sp. z o.o. in Kraków, Poland, under Polish law. The repository is young and the version is still 0.4.x, so expect change. Check the releases page for the latest version before you rely on it.

How does Nativ handle dictation audio and permissions?

According to the README, raw dictation audio is deleted after five minutes, and transcripts can be exported to PDF, DOCX or TXT. Microphone, accessibility and screen-recording permissions are optional, and the FAQ says you can skip any of them. PromptQuorum has not checked this.

Verdict

Nativ is worth a look if you own an Apple Silicon Mac and want a free, open-source local AI app that bundles chat, images, dictation and a local API for coding agents.

On paper it covers a lot: a bundled MLX server, a Hugging Face model library with fit warnings, OpenAI- and Anthropic-compatible endpoints, set-up for ten coding agents and six editors, an audio extension and MCP support, with v0.4.6 released on 16 November 2026 and a commit on 17 November 2026.

The trade-offs are real. It runs only on Apple Silicon with macOS 26 or newer, it is a young project with a 0.4 version number, and when PromptQuorum checked on 18 November 2026 the privacy policy contained a bracketed placeholder and sections on accounts and purchases. PromptQuorum has not tested the app, so start with a small model, read the current privacy policy and check that dictation, image generation and the API work for your use before relying on it.

Sources

  • Nativ website — features, requirements and downloads, checked 18 November 2026.
  • FAQ — requirements, privacy and integrations, last updated 5 November 2026, checked 18 November 2026.
  • Privacy policy — data handling, last updated 16 November 2026, checked 18 November 2026.
  • Terms of use — provider, licence and paid-plan wording, checked 18 November 2026.
  • GitHub repository — licence, creation date and last commit, checked 18 November 2026.
  • README — features, install, requirements and local API, checked 18 November 2026.
  • Release v0.4.6 — latest release, its date and the DMG, checked 18 November 2026.
  • Homebrew cask — cask version, checked 18 November 2026.

← Back to Power Local LLM

MCP