Home Page | Documentation | Discord

Private. Open source. One-click install. Developer-friendly API.

Install in one line:

curl -fsSL https://raw.githubusercontent.com/AtomicBot-ai/atomic-agent/main/scripts/install.sh | sh

A local operator agent for llama.cpp — same product class as OpenClaw Operator and Hermes Agent, but shipped as a standalone SEA binary and tuned to squeeze the most out of small local models. Your data, traces, browser profile, memory, and model traffic stay on your machine by default.

• 🧊 Built for local inference — stable cache-hot prompt prefix, externalized state in SQLite, GBNF grammar-constrained tool calls, parallel tool batches
• 🌐 Operates a real desktop — system browser via compact ARIA snapshots, shell, filesystem, documents (PDF/DOCX/XLSX/etc.), git, clipboard, HTTP, notifications
• 🧰 Skills, memory, tasks — local Markdown playbooks loaded on demand, FTS5 note recall, durable cron and webhook-triggered work
• 🔌 Multiple surfaces — TUI, CLI, OpenAI-compatible HTTP server, and a Tauri sidecar speaking newline-delimited JSON for desktop integrations
• 🛡️ Approvals & traces — dangerous actions gated by policy, append-only NDJSON traces with prompt-drift replay
• 💸 No SaaS meter — once you have local hardware and model files, no per-prompt, per-token, or per-seat fees

The easiest way to run OpenClaw. A native desktop app that turns the open-source AI agent framework (330k+ stars) into a personal AI assistant — no terminal, no config files, no Docker. Download, open, pick your AI provider, and start working.

Atomic Bot connects the best AI models to your everyday tools and actually takes action: drafts emails, schedules meetings, summarizes docs, automates browser tasks, and runs 13,000+ skills from ClawHub.

• 🔒 Private — runs locally, your data stays on your device
• 🆓 Free — use your own LLM API keys and pay nothing
• 🤖 Multi-model — Claude, GPT, Gemini, and more — switch on the fly
• 💬 Multi-messenger — one AI across Telegram, Slack, Discord, WhatsApp
• 🎙️ Voice — built-in Whisper transcription, local or cloud
• 🧠 Memory — persistent context across sessions and tasks
• 🔄 Auto-updates — always on the latest OpenClaw release

The agent that grows with you. A native AI assistant — not a browser tab, not a CLI wrapper. An autonomous agent with hands, eyes, memory, and a real workspace. Powered by the Hermes Agent core by Nous Research and tuned for self-improvement, time-travel file history, and fully offline operation.

• 👁️ Computer Use that actually lands the click — every screenshot is paired with native OCR (Apple Vision / Windows.Media.Ocr) so the agent gets pixel-accurate coordinates instead of guessing
• 🕰️ Time-travel file history — every file the agent touches is silently snapshotted before and after; diff or restore any version with one click, no git required
• 🧠 Self-improving skills & memory — the agent writes its own procedures after complex tasks and decides what's worth remembering across sessions
• ☁️ Local or cloud, one click — bundled inference engine downloads a model that fits your hardware, or connect 20+ cloud providers (OpenRouter, Anthropic, OpenAI, Gemini, DeepSeek, and more)
• 💬 One agent, 16+ messengers — Telegram, Discord, Slack, WhatsApp, Signal, iMessage, Email, Matrix, Teams, and others — all sharing the same memory
• 🛠️ 40+ tools, MCP-native — file ops, web search, code execution, subagents, cron scheduling, browser automation, Skills Hub from agentskills.io
• 🛡️ Approval gates — native modals confirm dangerous shell commands and writes before they run

Open-source ChatGPT alternative. Run local LLMs or connect cloud models — with full control and privacy. Download and run models from HuggingFace, or connect to OpenAI, Anthropic, Mistral, Groq, and others. Available on macOS, Windows, and iOS.

• 🧠 Local AI Models — download and run LLMs (Llama, Gemma, Qwen, and more) from HuggingFace
• ☁️ Cloud Integration — connect to OpenAI, Anthropic, Mistral, Groq, MiniMax, and others
• 🤖 Custom Assistants — create specialized AI assistants for your tasks
• 🔌 OpenAI-Compatible API — local server at localhost:1337 for other applications
• 🔗 Model Context Protocol — MCP integration for agentic capabilities
• 📱 Cross-platform — desktop apps for macOS and Windows, plus a native iOS app
• 🔒 Privacy First — everything runs locally when you want it to

Framework-agnostic desktop automation for AI agents. Give your agent eyes and hands — screenshot the screen, click, type, scroll, drag, read text via OCR, and more. Works with any tool-calling LLM framework, MCP server, or custom pipeline.

The library that powers computer use in Atomic Hermes — also published standalone for anyone to use:

• 📦 @atomicbotai/computer-use — the core TypeScript library: OCR, actions, overlay, session lock
• 🔌 @atomicbotai/computer-use-mcp — MCP server, drop it into Claude Desktop, Cursor, Windsurf, or any MCP client

Key features:

• 🔍 Zero-dependency OCR — native engines (Apple Vision on macOS, Windows.Media.Ocr on Windows) extract text with pixel-precise coordinates and confidence scores. No Tesseract, no cloud, no API keys
• 📸 Screenshot + anchors — turns a blurry downscaled screenshot into a structured map of the UI: "Send" at (1450, 890) instead of guessing where the button is
• 🖱️ Full action set — click / double / triple, type, press, scroll, drag, hold key, clipboard, open/switch app, list displays
• 🟢 Native overlay — Swift on macOS, PowerShell on Windows — visual indicator whenever the agent is driving the mouse and keyboard
• 🔒 Session lock — file-based lock prevents two agents from fighting over the desktop
• 🛡️ Guardrails — prevents misclicks in dock/launcher and submit zones
• 🐛 Debug artifacts — save screenshots, OCR results, and tool outputs per action for inspection

The official ClawHub API is limited — many endpoints are missing, and getting the full dataset means querying the Convex database directly.

ClawHub Layer API solves this by providing a complete, standalone REST API that aggregates everything into clean, ready-to-use endpoints:

• 🗂️ Full data coverage — skills, metadata, and everything the official API doesn't expose
• 🔄 Auto-syncing — periodically pulls and caches the latest data from Convex, always up to date
• 🛠️ Standard REST — clean endpoints, no Convex knowledge required
• ⚡ Fast & self-contained — no direct dependency on upstream infrastructure at query time
• 🔗 Integration-ready — designed to plug into your apps, bots, and workflows

Building on top of the OpenClaw / ClawHub ecosystem? This API is the simplest way to get the data you need — without going through Convex yourself.

We're building in the open and would love your help!

_{© 2026 Atomic · Built with ❤️ · atomicbot.ai}