AI agents that run natively and offline
IonClaw is a C++ AI agent orchestrator compiled natively for servers, desktop, Raspberry Pi, and mobile. It runs fully offline and requires zero external dependencies.
Native on every platform
A single C++ codebase compiles to a native build for Linux, macOS, Windows, mobile, and Raspberry Pi. It runs without any runtime, container, or external dependency.
The only one that runs on mobile
It turns your iPhone, iPad, Apple TV, Apple Watch, or Android device into a true personal AI assistant. Everything runs directly on your device, so it stays private and secure by design.
Offline and private
You can run local GGUF models through the built-in llama.cpp engine. There is no cloud and no API key, so your data and conversations never leave the device.
Your agents, everywhere you are
Run your agents on phones, tablets, watches, and TVs, and talk to them straight from WhatsApp and Telegram. Send a message from your chat app and make something happen on the device in your hand, on your wrist, or in your living room.
iPhone & iPad
Create, run, and monitor your agents from the native app on iOS and iPadOS.
Android
The same native app runs on Android phones and tablets, with the identical C++ core.
Apple Watch
Start and follow the local server from your wrist, and dictate a message to your agent.
Apple TV
The native tvOS app puts server controls on the big screen, with voice input and a QR code to reach the panel.
Message your agents on WhatsApp through Z-API or the official Meta Cloud API.
Telegram
Connect a bot token and chat with your agents from any Telegram client.
Desktop and servers run the same build too. The download page has every platform.
Quick Start
Clone, build, and run. The result is a single native binary that starts with one command.
# clone the repository $ git clone https://github.com/ionclaw-org/ionclaw.git $ cd ionclaw # build the native binary $ make setup-web $ make build-web $ make build $ make install # initialize a project and start the server $ ionclaw-server init ./my-project $ ionclaw-server start --project ./my-project → web panel ready at http://localhost:8080
Building needs CMake 3.25+, a C++20 compiler, Perl for the OpenSSL build (plus NASM on Windows), and Node.js 20.19+ for the web client. Every C++ dependency is fetched and built from source by CPM, so the compiled binary runs with zero runtime dependencies.
A complete platform, not just an agent
IonClaw brings together multi-agent orchestration, a browser-based control panel, a skills extension system, multi-provider LLM support, and WhatsApp and Telegram channels. It also ships with full MCP client and server integration inside a single native build.
Explore all featuresCommunity
Join the discussions and share your ideas. Contribute to the project and help shape where IonClaw goes next.
Free and open source
IonClaw is released under the MIT license. Read every line, run it wherever you want, change what you need, and ship it inside your own products, commercial ones included. There is no edition to buy, no account to create, and no telemetry sending your work somewhere else.
- The whole platform under the MIT license, with no features held back for a paid tier
- Free apps for iPhone, iPad, Apple TV, Apple Watch, and Android on the App Store and Google Play
- Local GGUF models that run offline on your own hardware at no cost
- Bring your own key for cloud providers and pay them directly, with no markup from IonClaw
- Self-hosted from day one, so your agents and conversations stay on machines you control
The only thing you can ever pay for is usage from a cloud LLM provider you choose to connect, and that bill goes straight to them. Run local models and there is nothing to pay at all.