autoMate

autoMate is an open-source, self-hosted backend layer for AI workflows, built by GitHub user yuruotong1 and released under the MIT license. It runs locally on your machine and exposes 40+ tools — notes, files, reminders, memory, shell execution, browser automation, and 31 SaaS integrations — to any AI client that supports MCP or HTTP. The project has accumulated over 3,900 GitHub stars and 489 forks since its creation in January 2024.

What It Is

autoMate describes itself as "a smart NAS for AI." Rather than being a chat interface itself, it is the persistent data and tool layer that sits behind whichever AI client you prefer — OpenClaw, Claude Desktop, Cursor, Cline, Kimi, or your own scripts. The core idea is that AI vendors provide chat and reasoning, but none of them persist memory across vendors, store your files, or send push notifications. autoMate fills that gap: it stores everything in ~/.automate/ using SQLite and Fernet encryption, and any connected client can read and write to the same warehouse.

How It Works

autoMate runs a FastAPI server on 127.0.0.1:8765 and exposes tools over MCP-over-HTTP. Connecting a client is a one-step process: open Settings → Connect to AI clients, click "Copy install text," and paste the resulting markdown blob into your client's config or hand it to another AI to configure automatically. Once connected, the client gains access to all autoMate tools including search.find, notes.read, files.list, audio.transcribe, and a top-level automate tool that runs autoMate's own agent loop on demand.

Two usage modes are supported:

• As a tool source — your preferred AI client (Claude Desktop, Cursor, etc.) acts as the brain and calls autoMate's tools for data and execution.
• Standalone web chat — autoMate's own built-in agent loop handles queries directly via its PWA frontend.

What's in the Box

The tool catalog spans several categories:

• Personal data: markdown notes with BM25 full-text search, a content-addressed file vault with deduplication, a reminder scheduler with Web Push to mobile, and long-term key-value memory readable by any AI.
• Local executors: shell commands, Python/Bash/Node scripts, desktop automation via pyautogui, Playwright browser control, and real-Chrome automation via a Chrome MV3 extension.
• SaaS integrations: 31 platforms including GitHub, GitLab, Notion, Slack, Linear, Jira, Confluence, Trello, Asana, Monday.com, HubSpot, Airtable, Stripe, Shopify, Telegram, Discord, MS Teams, Zoom, Twitter/X, SendGrid, Mailchimp, Twilio, Sentry, and several Chinese platforms (飞书, 钉钉, 企业微信, 微信公众号, 微博, 语雀, 高德地图).
• LLM providers: 25 providers in the catalog, including OpenAI, Anthropic, Gemini, xAI Grok, Mistral, DeepSeek, Ollama, LM Studio, and any OpenAI-compatible endpoint.

Deployment and Installation

autoMate supports four install paths:

• pip install automate-hub for Python users
• Standalone binaries for Windows, macOS, and Linux (no Python required)
• Docker via ghcr.io/yuruotong1/automate:latest for headless or NAS deployments
• An Android APK for a mobile viewer

The server binds to 127.0.0.1 by default; network access requires an explicit --host 0.0.0.0 flag. API keys and OAuth tokens are encrypted with Fernet, and the MCP endpoint requires a Bearer token.

Update: v4.6.0

The latest release is v4.6.0, published April 28, 2026. The README documents v4.5.7 as the current stable feature set, which introduced MCP-over-HTTP tool source mode, one-click "Copy install text" for AI client setup, Coze-style hybrid BM25 retrieval, audio transcription with custom vocabulary, configurable storage paths, and in-app auto-update. A paid autoMate Cloud Pro tier is described as upcoming, with the open-client/closed-server boundary documented in docs/cloud.md.