Features
IonClaw packs everything into a single native build. You get orchestration, a control panel, local models, WhatsApp and Telegram channels, and full MCP integration.
Key Capabilities
IonClaw is built in C++ for maximum performance and portability across every platform.
Real-Time Dashboard
The built-in web panel lets you monitor agents and track tasks in real time. You can control executions and inspect logs from the same place.
Skills & Marketplace
Skills written in plain Markdown extend what your agents can do. You add, edit, and manage them from the web dashboard, and the built-in marketplace installs community skills per project or per agent.
Multi-Provider LLM
IonClaw supports Anthropic, OpenAI, Gemini, Grok, OpenRouter, DeepSeek, and more. You pick models with a simple provider/model format, and credentials are stored by name so you can swap providers without rewriting config.
Local Models with llama.cpp
You can run GGUF models locally through the built-in llama.cpp engine on Linux, macOS, Windows, iOS, and Android. Everything stays fully offline and private. There is no cloud, no API key, and no data ever leaves your device.
Browser Automation
Agents drive a real Chrome through the DevTools Protocol. They navigate, fill forms, click, evaluate scripts, capture screenshots and PDFs, and inspect network traffic. It needs Chrome installed and runs on Linux, macOS, and Windows.
Task Tracking
Visual task management gives you real-time status updates. You can see what every agent is doing with full transparency and control.
Sandboxed Workspaces
Each agent runs inside an isolated workspace with controlled permissions. You whitelist tools per agent, set allow and deny policies, and add hooks, so nothing runs outside what you approved.
Agent-to-Agent Routing
Messages are routed intelligently between agents while preserving context. This enables structured collaboration between specialized instances.
Memory & Continuity
Persistent memory and session handling keep context across interactions. Your conversations and tasks stay continuous over time.
WhatsApp & Telegram
Your agents are reachable from the messaging apps you already use. WhatsApp works through Z-API or the official Meta Cloud API, and Telegram through a bot token. Each conversation keeps its own session.
Built-in Web Chat
Talk to your agents straight from the panel, with no external app. Messages stream in real time over WebSocket with media, tool use, and typing indicators. Sessions are saved so you can pick a conversation back up.
Native Tool Set
Agents get sandboxed file operations, a full HTTP client, web search, web fetch, and an RSS reader, plus shell execution on desktop and servers. Everything is compiled into the binary, with no external service required.
Images & Vision
Generate images through Gemini, OpenAI, or Grok, edit them locally with resize, draw, overlay, and watermark, and analyze any image with the model's native vision.
Subagents & Scheduling
An agent can spawn subagents to split up work and schedule recurring jobs with built-in cron. Automations keep running without anyone watching.
File Browser & Visual Config
The panel ships with a file browser that separates public and private files, plus a visual configuration editor with YAML validation. Secrets stay masked in the UI and are preserved when you save.
Device Notifications
Agents reach the host device through the platform bridge and can fire local notifications on iOS, watchOS, Android, macOS, and Linux.
MCP Client & Server
IonClaw fully supports the Model Context Protocol. It acts as both client and server, so it integrates with any MCP-compatible tool or platform.
Where It Runs
A single codebase compiles natively for every target. It reaches from cloud and desktop to iPhone, iPad, Apple TV, Apple Watch, and Android, down to Raspberry Pi and embedded hardware.
Cloud & Servers
You can deploy on any cloud provider, VPS, or dedicated server. It supports Docker and one-click deployment on Render, Heroku, DigitalOcean, and Hostinger.
Desktop
The same codebase produces native builds for Linux, macOS, and Windows. It needs no container, no virtual machine, and no runtime dependency.
Mobile, TV & Watch
Native apps run on iPhone, iPad, Apple TV, Apple Watch, and Android. IonClaw is the only AI agent orchestrator that goes this far across your devices. Everything runs on the device, so privacy and security come by design.
Raspberry Pi
You can run AI agents directly on Raspberry Pi boards. They handle edge computing, home automation, robotics, and local AI processing without any cloud dependency.
Embedded Systems
IonClaw is designed for resource-constrained environments. The same C++ codebase compiles for industrial controllers, edge gateways, and custom embedded hardware.
System Requirements
IonClaw itself is lightweight. When you run local GGUF models, the memory you need depends on the model size and quantization.
| Local model size | Approx. memory (4-bit) | Runs comfortably on |
|---|---|---|
| 1B – 3B | ~2 – 3 GB | Raspberry Pi 4/5, modern phones, any desktop |
| 7B – 8B | ~5 – 6 GB | 8 GB phones, Raspberry Pi 5 (8 GB), desktops and servers |
| 13B – 14B | ~9 – 10 GB | High-memory desktops and servers |
| 30B and above | ~20 GB or more | Workstations and servers, ideally with a GPU |
Figures are approximate for 4-bit quantization. On mobile and Raspberry Pi the model runs on the CPU, so available RAM is the limit. On desktops and servers, llama.cpp can offload layers to the GPU for more speed. Using a cloud provider such as Anthropic, OpenAI, or Google needs no local model memory at all.
Who It's For
IonClaw is built for anyone who needs AI agent orchestration. That ranges from embedded engineers to cloud architects.
C++ & Embedded Developers
You can build and extend the platform with C++. Create custom skills and deploy agents on the hardware you already work with.
IoT & Robotics Engineers
Run intelligent agents on edge devices, sensors, and robots. They process data locally without any cloud round-trip.
Teams & Organizations
Coordinate multiple agents across your infrastructure with structured workflows. You keep real-time visibility and centralized control.
Automation Enthusiasts
Design and deploy automated AI workflows on any hardware. They run anywhere from a Raspberry Pi at home to a production server in the cloud.
Production Environments
IonClaw is built for production with native performance and zero external dependencies. A single codebase simplifies deployment at scale.
Why IonClaw
These are the things that set IonClaw apart from other AI agent platforms.
Native Builds, Zero Dependencies
There is no Python, no Node.js, and no Docker. The platform compiles to a native build for your target. You just download it and run.
Native C++ Performance
IonClaw is built entirely in C++ for maximum efficiency. Its small memory footprint and fast execution matter on embedded and resource-constrained devices.
True Cross-Platform
One codebase compiles natively for Linux, macOS, Windows, iOS, watchOS, tvOS, Android, and Raspberry Pi. There are no platform-specific forks and no extra runtimes.
Reachable From Everywhere
Your agents live on iPhone, iPad, Apple TV, Apple Watch, and Android, and answer directly in WhatsApp and Telegram. Send a message from a chat app and trigger an action on any of your devices, all from the same platform.
Structured Orchestration
IonClaw is not a standalone experimental agent. It is a structured system that manages multiple agents with independent models, tools, and workspaces. MCP client and server support brings interoperability with external tools and platforms.
Offline & Private by Design
You can run GGUF models locally with built-in llama.cpp, fully offline and private. Your data and conversations never leave the device. There are no cloud round-trips and no API keys required.
Production-Ready
IonClaw is designed for real deployment with security, isolation, and operational transparency. It scales from edge devices to enterprise servers.
Frequently Asked Questions
Here are quick answers to the most common questions about IonClaw.
IonClaw runs on Linux, macOS, Windows, iOS, watchOS, tvOS, Android, Raspberry Pi, and embedded systems. Every target is compiled natively from a single C++ codebase.
Yes, it works completely offline. With built-in llama.cpp you can run local GGUF models on the device. There is no cloud, no API key, and no data ever leaves your device.
IonClaw supports Anthropic, OpenAI, Google, and other providers, along with local models. It is MCP-compatible as both client and server, so it integrates with any MCP tool or platform.
Yes, and it is the only AI agent orchestrator that runs natively on mobile. It turns your smartphone into a private personal assistant that works entirely on-device.
Yes, IonClaw is fully open-source under the MIT license. Contributions, feedback, and collaboration are always welcome.
Yes. IonClaw ships with WhatsApp and Telegram channels. WhatsApp connects through Z-API or the official Meta Cloud API, and Telegram through a bot token. You enable and configure both from the web panel, and each conversation keeps its own session and history.
On Apple, the app runs on iPhone, iPad, Apple TV, and Apple Watch. It is a single App Store download built from the same C++ core, and it requires iOS or tvOS 16 and watchOS 9 or later. On Android, the same app runs on phones and tablets from Android 7. Local GGUF models run on iPhone, iPad, and Android; on Apple TV and Apple Watch you connect to a provider or to a server running elsewhere.