AX Studio

The Open-Source AI Desktop Application

AX Studio is a native desktop app that connects any AI provider — cloud or self-hosted — through a single unified interface with MCP tools, local inference, and knowledge base integration.

Status: Open Source (Apache-2.0) | Tauri 2 + React 19 | macOS · Windows · Linux | v0.92

What AX Studio does: turns cloud and local AI providers into a production-ready desktop experience with conversation management, local inference, MCP integrations, artifact rendering, and a multi-agent framework.

Why teams use it: same chat interface, but with your own inference backend, your own retrieval layer, and full control over where your data lives.

---

Quick Start (60 Seconds)

Prerequisites

• Node.js ≥ 20
• Yarn ≥ 4.5.3
• Rust ≥ 1.77.2
• Tauri CLI ≥ 2.7.0

cargo install tauri-cli

Clone and Run

git clone https://github.com/defai-digital/ax-studio
cd ax-studio
make dev

make dev installs all dependencies, builds the core library and extensions, and launches the app with hot reload.

Connect a Provider

On first launch, go to Settings → AI Providers and add an API key for any supported provider, or point the app at a self-hosted backend.

---

Why AX Studio

Most AI desktop apps are thin wrappers around a single provider. AX Studio focuses on what happens around inference.

• Unified interface across 10+ cloud providers and any OpenAI-compatible endpoint
• First-class local inference via llama.cpp and ax-serving
• MCP (Model Context Protocol) client built in — connect tools, APIs, and databases
• Multi-agent framework with agent teams and execution logs
• Artifacts engine — render HTML, React, SVG, Chart.js, Vega-Lite inline
• Python code execution in a Docker sandbox
• Deep research engine with free web search
• Split-screen chat — two conversations side by side
• All data stays on your machine; cloud calls go direct from your device

---

Core Capabilities

Capability	AX Studio
Multi-provider chat (OpenAI, Anthropic, Mistral, Groq, Gemini, Azure, OpenRouter, HuggingFace)	✅
Local inference via llama.cpp + ax-serving	✅
MCP server integration (stdio + HTTP SSE)	✅
Multi-agent framework with agent teams	✅
Artifacts engine (HTML / React / SVG / Chart.js / Vega-Lite)	✅
Python code execution (Docker sandbox)	✅
Deep research engine with web search	✅
Mermaid diagram rendering	✅
Split-screen chat	✅
Semantic memory search	✅
Conversation threads with persistent history	✅
Custom system prompts per thread	✅
Model catalog — browse and download GGUF models from HuggingFace	✅
Voice — STT + TTS	✅
Local OpenAI-compatible API on localhost:1337	✅
TypeScript extension system	✅
Cross-platform (macOS, Windows, Linux)	✅

---

Supported Providers

Provider	Type
OpenAI	Cloud
Anthropic	Cloud
Azure OpenAI	Cloud
Mistral	Cloud
Groq	Cloud
Google Gemini	Cloud
OpenRouter	Aggregator
HuggingFace	Cloud + Hub
ax-serving / llama.cpp	Self-hosted
Any OpenAI-compatible endpoint	Self-hosted

---

---

MCP (Model Context Protocol)

AX Studio has a built-in MCP client. Add any MCP server in Settings → MCP Servers.

Supported transports:

• stdio — Child process with stdin/stdout
• HTTP SSE — Remote server with Server-Sent Events

Connected tools appear in the chat interface and can be toggled per-thread.

---

Local Inference

AX Studio supports running models entirely on your machine via two paths:

llama.cpp extension — bundled extension that manages llama-server processes and GGUF model loading directly from the app.

ax-serving — connect to a running ax-serving instance for production-style local inference with queuing, health-aware routing, and runtime load/unload.

# Point AX Studio at a local ax-serving instance
Settings → AI Providers → Ax-Serving → http://127.0.0.1:18080

---

Build from Source

Make Targets

Target	Description
make dev	Install deps + launch dev build with hot reload
make build	Production build for current platform
make test	Run tests and linting
make clean	Delete all build artifacts
make lint	Run ESLint
make dev-web-app	Frontend-only dev server (no Tauri)
make dev-android	Android development build
make dev-ios	iOS development build (macOS only)

Manual Steps

yarn install
yarn build:tauri:plugin:api    # Build Tauri plugin bindings
yarn build:core                # Build @ax-studio/core
yarn build:extensions          # Bundle extensions
yarn dev:tauri                 # Launch dev server

Platform Production Builds

yarn build:tauri:darwin    # macOS universal binary (.dmg)
yarn build:tauri:win32     # Windows installer (.exe)
yarn build:tauri:linux     # Linux packages (.deb + .AppImage)

---

Installation

Download the latest release from GitHub Releases:

Platform	Format
macOS (Universal)	.dmg
Windows	.exe installer
Linux (Debian/Ubuntu)	.deb
Linux (All distros)	.AppImage

---

Architecture

AX Studio is a Tauri 2 application: a React frontend embedded in a native Rust host.

┌──────────────────────────────────────────────────────────┐
│                     Desktop Window                       │
│                                                          │
│  ┌────────────────────────────────────────────────────┐  │
│  │            React Frontend  (TypeScript)            │  │
│  │                                                    │  │
│  │  Vercel AI SDK ── ModelFactory ── Providers        │  │
│  │  Zustand stores ── ServiceHub ── TanStack Router   │  │
│  │  Extension system ── MCP client ── i18n            │  │
│  └─────────────────────┬──────────────────────────────┘  │
│                        │  Tauri IPC                       │
│  ┌─────────────────────▼──────────────────────────────┐  │
│  │            Rust Backend  (Tauri + Tokio)            │  │
│  │                                                    │  │
│  │  File system ── Thread storage ── Download mgr     │  │
│  │  MCP server manager (rmcp) ── Provider configs     │  │
│  │  Local API proxy ── App updater ── Extension loader│  │
│  └────────────────────────────────────────────────────┘  │
└──────────────────────────────────────────────────────────┘
         │                              │
         ▼                              ▼
  Cloud AI Providers          Self-Hosted Backend
  (OpenAI, Anthropic,         (ax-serving, Retrieval,
   Mistral, Groq, etc.)        Agents, AkiDB)

---

Repository Layout

• web-app/ — React 19 frontend (Vite + TanStack Router)
• src-tauri/ — Rust backend: IPC commands, file I/O, MCP, downloads, local API proxy
• core/ — @ax-studio/core: extension interfaces and type definitions
• extensions/ — Bundled extensions: assistant, conversation, download, llama.cpp
• scripts/ — Build and CI scripts
• Makefile — Top-level build orchestration

---

Tech Stack

Frontend: React 19, TypeScript, Vite, TanStack Router, Zustand 5, Vercel AI SDK, Radix UI, Tailwind CSS 4, Vitest

Backend: Tauri 2.8, Tokio, rmcp 0.8 (MCP), Hyper/Reqwest, Serde

Tauri Plugins: tauri-plugin-http, tauri-plugin-store, tauri-plugin-shell, tauri-plugin-os, tauri-plugin-opener, tauri-plugin-updater, tauri-plugin-deep-link, tauri-plugin-log, tauri-plugin-single-instance, tauri-plugin-hardware (custom)

---

System Requirements

Platform	Minimum
macOS	13.6+ (Apple Silicon or Intel)
Windows	Windows 10+
Linux	glibc 2.31+

RAM requirements depend on model size. Cloud-only usage requires minimal resources.

---

Contributing

Contributions are welcome. Open an issue first for significant changes.

See CONTRIBUTING.md for coding standards, branch conventions, and the pull request process.

---

License

Apache 2.0