JMart Studio
v0.3.0-preview.1 — Early Preview

Download
JMart Studio

A customizable AI workspace that adapts to how you work — start with deep code intelligence, then shape it with the plugins, tools, and integrations you need. One assistant. Endless possibilities.

↓ Download for macOS How to get started

macOS Apple Silicon  ·  Windows x64  ·  Linux x64  —  all builds below

JMart Studio

v0.3.0-preview.1
JMart.Studio_0.3.0-preview.1_aarch64.dmg macOS · Apple Silicon (M1+) · 124 MB · Signed & notarized
Download
JMart.Studio_0.3.0-preview.1_x64-setup.exe Windows · x64 · 71 MB · NSIS installer
Download
JMart.Studio_0.3.0-preview.1_amd64.deb Linux · x64 · 129 MB · Debian package
Download
Setup
Get started in minutes

No Docker, no cloud accounts, no API keys required to get running.

1

Install JMart Studio

Available on macOS, Windows, and Linux.

2

Activate your license

Open Settings → License, paste your preview key, and click Activate. Don't have a key yet? Request one below — it's free for 30 days.

3

Connect an AI model

Go to Settings → AI Providers. Add Anthropic, OpenAI, or point it at a local Ollama instance. JMart Studio works with any OpenAI-compatible endpoint.

4

Open a project and start building

Open a folder, start a chat, or launch the MCP server so Claude Code and Cursor can use your project context directly.

What's inside
Everything in one place

Built for developers who want AI as a native part of their workflow — not a browser tab.

🧠

Local AI Engine

Runs models fully on-device via Ollama, or connects to Anthropic and OpenAI. Your code never leaves your machine unless you explicitly choose a cloud model.

🔌

MCP Tool Server

Exposes a local MCP endpoint on port 3456. Claude Code, Cursor, and any MCP-compatible agent can use your project's live context — file tree, symbols, git state, terminal.

🗂️

Code Intelligence

Semantic file indexing, symbol inspection, and codebase search — all local, all fast. The AI always knows what's in your project without uploading it anywhere.

🎙️

Voice Input

Built-in Whisper STT for voice-to-text in chat. Speak your prompts, review them, send. Works fully offline using the local Whisper model.

🐳

Sandbox Mode

Run agent tasks inside an isolated container. Supports Docker and Apple Container — the AI can write and execute code without touching your host system.

🔒

Secrets Manager

Store API keys and credentials in the macOS Keychain — never in plaintext, never in .env files. Access them from your agents without exposing them in prompts.

🗃️

Kanban Board

Track tasks, agent runs, and project work in a built-in kanban board. Agents can create and move cards. Everything stays local in SQLite.

Bring Your Own Model

Works with Anthropic Claude, OpenAI GPT, Ollama local models, or any OpenAI-compatible API endpoint. Switch models per task — route fast tasks to cheap models automatically.

🔄

Auto-Updates

Built-in updater checks for new releases in the background. One click to install — signed binaries, verified on-device before applying.

Early Preview

Get your free preview license

JMart Studio is in private preview. Enter your email to receive a 30-day license key instantly.

FAQ
Common questions

Is JMart Studio free?

The preview is free for 30 days with a license key. After the preview period, JMart Studio will be a one-time purchase — no subscriptions for the core IDE. You bring your own AI API keys or run local models free of charge.

Does it work offline?

Yes, for local models (Ollama). The IDE, code intelligence, kanban, terminal, and secrets manager all work without an internet connection. Cloud AI features (Anthropic, OpenAI) require connectivity only for those API calls.

What AI models are supported?

Any model accessible via an OpenAI-compatible API — including Anthropic Claude, OpenAI GPT-4o, and local models via Ollama (Llama 3, Mistral, Qwen, and hundreds more). You can add multiple providers and route tasks between them.

Is my code sent to the cloud?

Only when you explicitly use a cloud AI model and include file context in a prompt. The codebase index, search, symbols, git state, and all local features never leave your machine. You are always in control of what gets sent.

Which platforms are supported?

macOS on Apple Silicon (M1 and later — signed & notarized), Windows x64 (NSIS installer), and Linux x64 (Debian .deb package) — all available above. macOS and Windows update themselves in-app; the Linux build is download-only for now (re-download to update).

Windows shows an "unknown publisher" warning?

This preview's Windows build isn't Authenticode code-signed yet, so SmartScreen flags it. Click More info → Run anyway to install — it's safe. Code signing is on the roadmap; the macOS build is already fully signed and notarized.

What is the MCP server?

JMart Studio runs a local Model Context Protocol server on port 3456. This lets Claude Code, Cursor, and other MCP-compatible AI agents access your project's live context — file tree, symbols, git status, terminal — directly from their own sessions.