Website · Docs · Discord · X

Open-source local AI — agents, chat, and inference. Private by default.

An open-source desktop AI app. Run local LLMs from Hugging Face or connect cloud models (OpenAI, Anthropic, Mistral, Groq, MiniMax, others). Available on macOS, Windows, Linux, iOS, and Android.

Desktop

Mobile

curl http://localhost:1337/v1/chat/completions -d '{
  "model": "llama-3.2-3b-instruct",
  "messages": [{ "role": "user", "content": "Why is the sky blue?" }]
}'

• Run open-weight LLMs (Llama, Gemma, Qwen, Mistral, Phi) from Hugging Face — fully offline
• Connect cloud providers: OpenAI, Anthropic, Mistral, Groq, MiniMax
• One-click agent launch — OpenCode & GitHub Copilot CLI from the Integrations tab
• Artifacts — live preview panel for HTML/CSS/JS
• Faster on Apple Silicon — MLX EAGLE-3, MTP, and TurboQuant KV cache
• MCP integration for agentic capabilities; custom assistants for specialized tasks
• Native iOS & Android apps, not wrappers

A llama.cpp fork with TurboQuant KV cache compression and Gemma 4 MTP speculative decoding. ~30-50% throughput gains on the same hardware, drop-in compatible with upstream tools and GGUF.

WHT-rotated 2/3/4-bit KV cache with backend-native kernels (Metal TurboFlash, CUDA, Vulkan, HIP). turbo3 is the default — 3-bit, ~4.3× compression vs F16.

llama-server -m model.gguf -c 32768 -ngl 99 -fa on \
  -ctk turbo3 -ctv turbo3

Pair any gemma4 target with the official gemma4_assistant head — loaded into the target context, no second tokenizer or KV cache. +30-50% short-prompt throughput on Gemma 4 26B-A4B / 31B at 85-88% accept rate. Pre-built assistant heads on Hugging Face.

llama-server -m gemma-4-target.gguf -c 16384 -ngl 99 -ngld 99 -fa on \
  --mtp-head gemma-4-assistant-Q4_K_M.gguf \
  --spec-type mtp --draft-block-size 3

• TQ3_1S / TQ4_1S weight quantization via llama-quantize — 25-35% smaller than Q8_0, single-digit % PPL delta
• Regularly synced with ggml-org/llama.cpp
• Powers local inference in Atomic Chat, Atomic Hermes, and Atomic Agent

A native autonomous AI agent for desktop. Built on the Hermes Agent core by Nous Research, with computer use, time-travel file history, and offline operation.

• Computer use with native OCR (Apple Vision / Windows.Media.Ocr) — pixel-accurate click coordinates, no guessing
• Time-travel file history: every file the agent touches is snapshotted before and after, one-click diff or restore
• Self-improving skills and memory: the agent writes its own procedures and decides what to remember across sessions
• Bundled inference engine, or 20+ cloud providers (OpenRouter, Anthropic, OpenAI, Gemini, DeepSeek, others)
• One agent across 16+ messengers — Telegram, Discord, Slack, WhatsApp, Signal, iMessage, Email, Matrix, Teams
• 40+ tools, MCP-native: file ops, web search, code execution, subagents, cron, browser automation, agentskills.io Skills Hub
• Approval modals for dangerous shell commands and writes

A native desktop app that turns OpenClaw (330k+ stars) into a personal AI assistant. No terminal, no config, no Docker.

• Drafts emails, schedules meetings, summarizes docs, automates the browser
• 13,000+ skills from ClawHub
• Multi-model: Claude, GPT, Gemini — switch on the fly with your own API keys
• One AI across Telegram, Slack, Discord, WhatsApp
• Built-in Whisper transcription, local or cloud
• Persistent memory across sessions and tasks
• Auto-updates to the latest OpenClaw release

Framework-agnostic desktop automation for AI agents. Screenshot, click, type, scroll, drag, OCR — works with any tool-calling LLM, MCP server, or custom pipeline.

npm install @atomicbotai/computer-use

import { screenshot, click, type } from "@atomicbotai/computer-use";

const { image, anchors } = await screenshot();
const send = anchors.find(a => a.text === "Send");
await click(send.x, send.y);
await type("hello");

• @atomicbotai/computer-use — TypeScript library: OCR, actions, overlay, session lock
• @atomicbotai/computer-use-mcp — MCP server for Claude Desktop, Cursor, Windsurf, or any MCP client

• Zero-dependency native OCR (Apple Vision on macOS, Windows.Media.Ocr on Windows) — no Tesseract, no cloud, no API keys
• Pixel-accurate UI anchors: "Send" at (1450, 890) instead of guessing from a downscaled screenshot
• Full action set: click / double / triple, type, press, scroll, drag, hold key, clipboard, app switch, list displays
• Native overlay (Swift on macOS, PowerShell on Windows) shows when the agent is driving the mouse and keyboard
• File-based session lock prevents two agents from fighting over the desktop
• Guardrails against misclicks in dock/launcher and submit zones
• Per-action debug artifacts: screenshots, OCR results, tool outputs

A complete REST API for ClawHub. The official ClawHub API exposes a subset of the data; this layer aggregates everything into clean endpoints, no Convex knowledge required.

• Full coverage: skills, metadata, and everything the official API doesn't expose
• Auto-syncing: periodically pulls and caches the latest data from Convex
• Standard REST, no upstream dependency at query time
• Drop-in for apps, bots, and workflows

A local-first operator agent for llama.cpp. Standalone SEA binary, tuned for small local models. Data, traces, browser profile, memory, and model traffic stay on your machine.

curl -fsSL https://api.atomicbot.ai/agent-install | sh

atomic-agent

• System browser via ARIA snapshots, shell, filesystem, documents (PDF/DOCX/XLSX), git, clipboard, HTTP, notifications
• GBNF grammar-constrained tool calls, parallel tool batches, cache-hot prompt prefix, externalized state in SQLite
• Local Markdown skills loaded on demand, FTS5 note recall, durable cron and webhook-triggered tasks
• TUI, CLI, OpenAI-compatible HTTP server, and a Tauri sidecar speaking newline-delimited JSON
• Policy-gated dangerous actions, append-only NDJSON traces with prompt-drift replay

_{© 2026 Atomic · atomicbot.ai}