NativeCode logo NativeCode
Download Download
Native macOS app · version 1.3.5

A local AI coding editor for your Mac.

NativeCode combines a native SwiftUI editor, Monaco, Rapid-MLX, and Qwen3 models into a private AI workspace. Project files, prompts, and outputs stay on your device.

Download NativeCode.dmg 8.06 MB Installation guide
System
macOS 14+
Hardware
Apple Silicon recommended
AI runtime
Rapid-MLX · localhost:8000
SHA-256 · macOS .dmg b97f148f23b4f53d70957b0103fa7594e5d598928f9ecf34a8f90cd42a5a84c6

NativeCode

EditorView.swift
1struct EditorView: View {
2  var body: some View {
3    MonacoWebView(appState: appState)
4      .background(Theme.bg)
5  }
6}
Native Windows app · beta 1.0.3

A local AI coding editor for Windows.

NativeCode for Windows is a ground-up native WinUI 3 app — not Electron — with the Monaco editor, a project-aware agent, and local models via Ollama. Your project stays on your PC.

Download for Windows Beta · 61.90 MB Quick start
System
Windows 10/11
Hardware
x64 · 16 GB+ RAM
AI runtime
Ollama · local or remote
SHA-256 · .exe e60258c6656a43a6df2d0d50a5c0acb18d93dbce1204020f936cf487f5cbcf6e

NativeCode

—□✕
game.js
1const bird = {
2  y: 240, vy: 0,
3  flap() { this.vy = -6; }
4};
100% local no cloud API and no account
WinUI 3 ground-up native app, not Electron
Ollama local models, or any OpenAI-compatible server
100% local no cloud API and no account
Qwen3 MLX Qwen3, Qwen3-Coder, Qwen3-VL and Gemma 3 models in 4/6/8-bit variants
Monaco full code editor embedded in a native app

Features

An IDE that can work with your project.

NativeCode is not just chat next to an editor. The AI receives safe project tools and can read, search, edit, and create files.
01

Native macOS editor

SwiftUI interface with a file tree, recent projects, image preview, and Monaco editor for code.

02

Project-aware AI agent

The agent uses tools like `read_file`, `edit_file`, `search_in_files`, `find_files`, and more directly inside the opened project.

03

Controlled terminal access

The AI can run shell commands through `run_terminal`, while destructive commands are blocked and writes require approval.

04

Slovak and English

The UI and system prompt are bilingual. Replies follow the user's language while code identifiers stay unchanged.

New in version 1.3.5

Your chat history, always one click away.

Version 1.3.5 adds chat history. Your conversations are now saved automatically, so you can pick up exactly where you left off. A new History button sits right next to New chat — open it to browse and reopen any of your past chats.

Highlights

  • Chat history. Conversations are now saved automatically and stay around between sessions, so nothing gets lost when you close a chat.
  • A new History button. It lives right next to New chat — one click opens your saved conversations to browse and reopen.

How to update

  • Download the new NativeCode.dmg above and drag NativeCode into Applications, replacing the old version. Settings, models, and projects stay in place.
  • The app is still unsigned, so macOS may warn again on the first launch of the new version — open it via Open Anyway as described in the install section below.

Notes

  • Versioning moved from 1.3.4 to 1.3.5 and the build number from 10 to 11.

Features

Agentic coding, native on Windows.

A true WinUI 3 app with the Monaco editor, a project-aware agent, and local models through Ollama — everything running on your PC.
01

Native WinUI 3 app

Fluent dark UI in a self-contained app — no .NET runtime install, not Electron, not a web wrapper.

02

Local & remote models

Run models locally via `Ollama`, or point at any OpenAI-compatible server like LM Studio — with GPU offload control.

03

Agentic tools & terminal

File read/edit/search, shell commands, visible task lists, plan mode with approval, and a real ConPTY terminal on `Ctrl-J`.

04

Monaco, vision & bilingual

VS Code-style Monaco editing in WebView2, image paste for vision models, and an English & Slovak UI.

New · Windows 1.0.3 beta

Chat history comes to Windows too.

Beta 1.0.3 brings chat history to the Windows app — the same feature the Mac app got in version 1.3.5. Conversations are now saved automatically, and a new History button next to New chat lets you browse and reopen any past chat.

Highlights

  • Chat history. Conversations are now saved automatically and stay around between sessions, so nothing gets lost when you close a chat.
  • A new History button. It lives right next to New chat — one click opens your saved conversations to browse and reopen.
  • Native WinUI 3 app. Fluent, dark-themed, and self-contained — no .NET runtime install required.
  • Local and remote models. Run models locally via Ollama, or point at any OpenAI-compatible server (LM Studio, a remote Ollama).
  • Private by default. With local models your project stays on your PC — no cloud, no account.
  • Monaco editor. The same VS Code-style editing, hosted in WebView2.
  • Agentic tools. File read/edit/search, shell commands, visible task lists, and plan mode with approval.
  • Integrated terminal. A bottom terminal panel (Ctrl-J) backed by a real ConPTY session.
  • Vision support. Paste images into chat when the selected model supports it.
  • Built-in model management. GPU offload control, network model discovery, one-click Ollama install, and an English & Slovak UI.

System requirements

  • Windows 10 (1809+) or Windows 11, 64-bit (x64).
  • 16 GB RAM minimum, 32 GB recommended for larger models.
  • Ollama for local models — installable from inside the app.
  • Around 5–15 GB of free disk space, depending on the selected model.

Notes

  • Versioning moved from 1.0.1-beta to 1.0.3-beta.

Local AI

Ollama runs the models on your PC.

Install Ollama in one click from inside the app, pull a model, and NativeCode talks to it over a local OpenAI-compatible API — or point it at a remote server just as easily.
> winget install Ollama.Ollama
> ollama pull qwen3:8b
> ollama serve
http://localhost:11434/v1 · OpenAI-compatible · streaming

Install

From installer to your first local prompt.

NativeCode installs per-user — no admin rights. Then install Ollama and pull a model right inside the app.
Download for Windows Beta · 61.90 MB
SHA-256 · .exe e60258c6656a43a6df2d0d50a5c0acb18d93dbce1204020f936cf487f5cbcf6e

Installing NativeCode on Windows

  • 1. Download NativeCode-1.0.3-beta-setup.exe with the button above.
  • 2. Run it — it is a per-user install, so no administrator rights are needed.
  • 3. Install Ollama when prompted (one click) to run models locally, or point the app at a remote OpenAI-compatible server.
  • 4. Open a project folder and start coding with your AI agent.

The beta is not code-signed yet, so Windows SmartScreen may show “Windows protected your PC”. Click More info → Run anyway to continue — this is expected for apps distributed outside the Microsoft Store.

Models

Models you can run through Ollama.

Pull any of these on demand from inside the app, or use any other Ollama tag. Larger models need more RAM and benefit from GPU offload.
Family Sizes Type Ollama
Qwen3 1.7B · 4B · 8B · 14B · 32B Thinking + code ollama pull qwen3
Qwen3-Coder-30B 30B (MoE) Coding ollama pull qwen3-coder
Qwen3-VL 8B · 30B Vision + code ollama pull qwen3-vl
Gemma 3 4B · 12B · 27B Multimodal ollama pull gemma3
Any Ollama tag — Your choice ollama pull <tag>

Privacy and control

Your code stays on your PC.

With local Ollama models nothing leaves your machine, and every file write or command runs behind an approval flow.
No cloud by default

With local models, chat and tools run against Ollama on your PC — no account, no API key, zero telemetry.

Approval before writes

Write tools, deletes, and terminal commands ask for confirmation or per-thread approval.

You can verify it

No telemetry by default — confirm it yourself with Windows Defender Firewall or GlassWire.

FAQ

Windows questions.

Do I need an internet connection?

Only to install Ollama and pull models. Once a model is downloaded, it runs locally on your PC.

How is this different from Cursor?

NativeCode is a true native WinUI 3 app, not an Electron wrapper. It uses far less memory and runs models locally through Ollama.

Which models can I run?

Any Ollama tag — Qwen3, Qwen3-Coder, Qwen3-VL, Gemma 3 and more. Pull them from inside the app, or point at a remote OpenAI-compatible server.

Is my code sent anywhere?

Not with local models. NativeCode has zero telemetry by default — verify it with Windows Defender Firewall or GlassWire.

Why does SmartScreen warn me?

The beta isn't code-signed yet. Click More info → Run anyway — expected for apps distributed outside the Microsoft Store.

Found a bug or want a feature?

Report it on GitHub or email info@jakubhecht.sk.

Screenshots

See NativeCode for Windows in action.

The native WinUI 3 build with the file tree, Monaco editor, AI panel, and plan mode — exactly as it runs on Windows 10 and 11.
NativeCode for Windows screenshot
NativeCode running on Windows 11 in plan mode.

Local AI

Rapid-MLX runs on your Mac.

The app manages an isolated Python venv in Application Support, installs Rapid-MLX, and starts an OpenAI-compatible API on `localhost:8000`.
$ python3 -m venv ~/Library/Application Support/NativeCode/rapid-mlx-venv
$ pip install -U rapid-mlx
$ rapid-mlx serve --port 8000 mlx-community/Qwen3-8B-4bit
HTTP /v1/chat/completions · streaming · tool calls

Install

From DMG to your first local prompt.

NativeCode installs like a normal macOS app. Then you prepare the AI runtime and models directly in the app settings.
  1. 1

    Download the DMG

    Use the Download button and open `NativeCode.dmg`.

  2. 2

    Move the app

    Drag NativeCode into Applications and launch it.

  3. 3

    Install Rapid-MLX

    In Settings → Rapid-MLX, click Install. The app creates its own venv and installs the server.

  4. 4

    Download a model and start the server

    In Settings → Models, choose a Qwen3 model, download it, then start the server in the Rapid-MLX section.

Opening NativeCode — it is an unsigned app

NativeCode is built without a paid Apple Developer account, so it is not signed or notarized. On the first launch macOS warns that it cannot verify the app is free of malware. That is expected for any app distributed outside the App Store — here is how to open it on the latest macOS (26 Tahoe, and 15 Sequoia).

  • 1. Open NativeCode.dmg and drag NativeCode into your Applications folder.
  • 2. Double-click NativeCode. A dialog says Apple cannot verify it — click Done (do not choose "Move to Trash").
  • 3. Open System Settings → Privacy & Security and scroll down to the Security section.
  • 4. Next to "NativeCode was blocked to protect your Mac" click Open Anyway (available for about an hour), then confirm with Touch ID or your password.
  • 5. Launch NativeCode again and in the new dialog click Open Anyway. From now on it opens with a normal double-click.

If macOS instead says "NativeCode is damaged and can't be opened", remove the quarantine flag in Terminal and then open the app — this is the fastest, most reliable method:

$ xattr -dr com.apple.quarantine /Applications/NativeCode.app

Models

Supported MLX models.

Models are downloaded with `hf download` and stored in `~/.cache/huggingface/hub`. NativeCode tracks progress by polling file size on disk.
Family Quantization Size Hugging Face
Qwen3-8B 4-bit 4.31 GB mlx-community/Qwen3-8B-4bit
Qwen3-8B 6-bit 6.21 GB mlx-community/Qwen3-8B-6bit
Qwen3-8B 8-bit 8.12 GB mlx-community/Qwen3-8B-8bit
Qwen3-14B 4-bit 7.75 GB mlx-community/Qwen3-14B-4bit
Qwen3-14B 6-bit 11.19 GB mlx-community/Qwen3-14B-6bit
Qwen3-32B 4-bit 17.18 GB mlx-community/Qwen3-32B-4bit
Qwen3-Coder-30B 4-bit (MoE) 16.02 GB mlx-community/Qwen3-Coder-30B-A3B-Instruct-4bit
Qwen3-Coder-30B 6-bit (MoE) 23.12 GB mlx-community/Qwen3-Coder-30B-A3B-Instruct-6bit
Qwen3-Coder-30B 8-bit (MoE) 30.23 GB mlx-community/Qwen3-Coder-30B-A3B-Instruct-8bit
Qwen3-VL-4B 4-bit 3.09 GB mlx-community/Qwen3-VL-4B-Instruct-4bit
Qwen3-VL-4B 8-bit 5.10 GB mlx-community/Qwen3-VL-4B-Instruct-8bit
Qwen3-VL-8B 4-bit 5.76 GB mlx-community/Qwen3-VL-8B-Instruct-4bit
Qwen3-VL-8B 6-bit 7.81 GB mlx-community/Qwen3-VL-8B-Instruct-6bit
Qwen3-VL-8B 8-bit 9.86 GB mlx-community/Qwen3-VL-8B-Instruct-8bit
Qwen3-VL-30B 4-bit (MoE) 18.25 GB mlx-community/Qwen3-VL-30B-A3B-Instruct-4bit
Gemma3-12B 4-bit QAT 8.03 GB mlx-community/gemma-3-12b-it-qat-4bit
Gemma3-27B 4-bit 16.83 GB mlx-community/gemma-3-27b-it-4bit

Privacy and control

Your code does not leave your Mac.

NativeCode uses a local server and an approval flow for operations that can change files or run commands.
No cloud API

Chat and tool calls go to `localhost:8000`; no OpenAI or Anthropic key is required.

Approval before writes

Write tools, deletes, and terminal commands ask for confirmation or per-thread approval.

Dangerous commands blocked

Patterns like `rm -rf`, `dd`, `mkfs`, `diskutil erase`, and `shutdown` are blocked even in auto-approve mode.

FAQ

Common questions.

Do I need internet?

Yes for downloading Rapid-MLX and models. Work with an already downloaded model runs locally.

Where are models stored?

In the Hugging Face cache: `~/.cache/huggingface/hub`.

Can the AI edit files?

Yes. It has tools for reading, editing, creating, and deleting files, but write operations require your approval.

What Mac do I need?

The app targets macOS 14+. For local AI through MLX, an Apple Silicon Mac with enough RAM for the selected model is recommended.

Is there a Windows version?

Yes — a native WinUI 3 port is in beta. It runs local models through Ollama or any OpenAI-compatible server. Grab the installer in the Windows section; the beta is not code-signed yet, so SmartScreen may warn — click More info → Run anyway.

I found a bug or want a feature. Where should I write?

Report it on GitHub or email info@jakubhecht.sk.

Screenshots

See NativeCode inside the app.

This current screenshot shows the native editor UI, file tree, AI panel, and the overall product layout as it looks in real use.
NativeCode app screenshot on macOS
Current NativeCode view on macOS.

Mac & Windows · Free

Download NativeCode and run a local AI IDE.

Download for macOS 8.06 MB Download for Windows Beta · 61.90 MB