yuruotong1/autoMate

Like Manus, Computer Use Agent(CUA) and Omniparser, we are computer-using agents.AI-driven local automation assistant that uses natural language to make computers work by themselves

autoMate – a personal “NAS” for AI assistants

What it is – autoMate is a locally‑hosted service that gives any LLM‑powered chat client a persistent personal data store and a toolbox of real‑world actions. It keeps your notes, files, reminders, and long‑term “memory” on your own machine (SQLite + encrypted blobs) and exposes them to AI agents via a simple MCP‑over‑HTTP protocol. The same backend can be used from a web UI, a Chrome extension, or plugged into other AI front‑ends such as OpenClaw, Claude Desktop, Cursor, Cline, Kimi, etc.

Core ideas

Concept What it does
Personal data layer • notes.* – markdown notes with tags and full‑text search
• files.* – content‑addressed blob vault (deduplication, configurable storage path)
• reminders.* – scheduler that pushes Web‑Push notifications to the PWA
• memory.* – key‑value store that any AI can read/write across sessions
• search.find – hybrid BM25 retrieval over notes + files (SQLite FTS5)
Tool toolbox Shell, script (Python/Bash/Node), desktop automation (pyautogui), browser automation (Playwright), and a Chrome‑extension bridge (bx.*) that lets the AI control your real Chrome instance
SaaS integrations 31 ready‑made connectors (GitHub, Notion, Slack, Jira, Trello, Stripe, etc.) that the AI can call as tools
LLM catalog 25 providers (OpenAI, Anthropic, Gemini, Groq, Ollama, etc.) – the AI client picks a model and autoMate forwards the request directly to the provider
MCP bridge A lightweight HTTP endpoint (/mcp/) that any client can call with a bearer token to invoke the above tools. The bridge translates the LLM’s “function call” JSON into local actions.

How it works (high‑level flow)

  1. Start the server – automate (or the binary/Docker container) launches a FastAPI app on 127.0.0.1:8765.
  2. Pick a model – the web wizard asks for an API key and lets you select one of the 25 providers.
  3. Connect a client – In the Settings UI you click Copy install text. This copies a markdown snippet containing the MCP URL and a bearer token, plus ready‑made config sections for OpenClaw, Claude Desktop, Cursor, Cline, etc.
  4. Client reads the snippet – you paste it into the client’s config (or ask the client’s own LLM to do it). The client now knows how to reach autoMate.
  5. Chat → tool call – When you ask the LLM something that requires a tool (e.g., “add a note about the meeting”, “search my project files”, “schedule a reminder for tomorrow”), the client sends a JSON‑RPC call to /mcp/.
  6. autoMate executes – The server runs the requested tool (shell command, file write, reminder push, SaaS API call, etc.), updates its SQLite store, and returns the result to the client.
  7. Result appears in the chat – The LLM can incorporate the output and continue the conversation, and the data is now persisted for any future client.

Installation options

Method What you get When to use
Python package (pip install automate-hub) Small, source‑level install; requires a Python runtime You already have Python and want to tweak the code
Standalone binary (Windows/macOS/Linux) Double‑click executable; no Python needed End‑users on a desktop who just want to run it
Docker (docker run -p 8765:8765 ghcr.io/yuruotong1/automate:latest) Containerised, headless‑friendly; good for NAS or servers Deploy on a headless box, CI, or inside a home‑lab
Chrome extension (extension/ folder) MV3 extension that lets the AI control your real Chrome tabs via bx.* When you need the AI to browse the web or interact with existing pages
PWA / Phone Open the local URL on mobile Chrome/Safari and add to Home Screen Quick access from a phone without installing an app

After starting, open http://127.0.0.1:8765 – the wizard walks you through model selection, API‑key entry, and optional client wiring.

Typical use cases

  • Cross‑client memory – Write a note in Claude Desktop, later retrieve it from Cursor or a phone PWA.
  • Personal knowledge base – Store project docs, code snippets, PDFs in files.*; search them with a single search.find call.
  • Reminders & alerts – Schedule a reminder from any chat; the server pushes a Web‑Push notification to your phone.
  • Automation scripts – Trigger shell scripts, run Python code, or control the desktop directly from a conversation.
  • SaaS workflow – Create a GitHub issue, update a Notion page, or send a Slack message without leaving the chat.
  • Hybrid retrieval – Ask the LLM “what did I write about the 2023 budget?” – autoMate runs a BM25 search over notes + files and returns the most relevant excerpts.

Privacy & security highlights

  • The server binds only to 127.0.0.1 by default; exposing it externally requires --host 0.0.0.0.
  • All secrets (API keys, OAuth tokens, push subscriptions) are encrypted with Fernet; the key lives in ~/.automate/secret.key (chmod 600).
  • LLM calls go straight from your machine to the chosen provider – no third‑party proxy.
  • The MCP endpoint requires a bearer token; treat it like a password because it grants access to all tools, including shell.exec.
  • An optional “autoMate Cloud” tier exists for paid users, but it is opt‑in; without the AUTOMATE_CLOUD_URL env var no data leaves your device.

Current status

  • Version 4.5.7 – stable release with full MCP‑over‑HTTP support, 31 SaaS connectors, 25 LLM providers, audio transcription (Tencent ASR / OpenAI Whisper), and an auto‑update mechanism.
  • Open‑source under the MIT license.

TL;DR

autoMate turns your laptop (or NAS) into a personal AI‑aware data hub. It stores notes, files, reminders, and arbitrary key‑value facts locally, offers a rich toolbox of executors, and exposes everything through a simple HTTP‑based tool‑call interface that any LLM client can plug into. Install it once, connect your favorite chat front‑end, and enjoy a persistent, private AI assistant that remembers you across sessions and vendors.

Related

  • Project
    24kchengYe/MemoMind

    MemoMind is a locally hosted, GPU‑accelerated memory system for AI coding agents. It stores extracted facts from chats, documents, and daily life events in PostgreSQL + pgvector, builds a knowledge graph, and offers fast 4‑way hybrid retrieval (semantic, BM25, graph, temporal). The agent can retain new information, recall relevant memories, and reflect across the whole store. All data stays on the user’s machine, works with any OpenAI‑compatible LLM, and includes a web dashboard for browsing and exporting memories.

  • Project
    TencentCloud/Octop

    Octop is an open‑source, self‑hosted AI assistant platform. It runs a single FastAPI‑based process that serves a React dashboard, a CLI, and bots for Feishu, DingTalk, QQ, Discord, WeCom, etc. Users can create multiple agents with distinct MBTI‑style personalities, each with its own persistent memory and configurable LLM provider (OpenAI‑compatible, DashScope, Ollama, …). All data stays locally under `~/.octop/`, and the system includes guardrails, JWT isolation, browser/terminal automation, and a bidirectional ACP protocol for IDE integration. Installation is via a one‑line script, PyPI, or Docker; the project is MIT‑licensed and actively maintained.

  • Dispatch
    Rowboat Open-Source Local-First AI Coworker – Features, Architecture, and Community Feedback

    Rowboat is an open‑source desktop AI coworker that stores all data locally as Markdown, builds a persistent knowledge graph, and provides built‑in work surfaces like email, notes, browser, and code agents.

  • Dispatch
    Kept: Building a Local-First Knowledge Base from AI Conversations

    Explore Kept, an open-source tool that archives AI chats into a local Markdown vault, enabling full-text search, graph visualization, and agentic access via MCP.

  • Project
    DeusData/codebase-memory-mcp

    A native, zero‑runtime executable that builds a persistent, multi‑language knowledge graph of a codebase (158 languages) using Tree‑sitter and a custom “Hybrid LSP” type resolver. The graph can be queried instantly (structural, semantic, Cypher‑like) and is consumed by AI coding agents (Claude Code, Codex, OpenCode, etc.) to answer code‑understanding questions without sending source files to the cloud. Includes a built‑in 3D UI, cross‑repo linking, dead‑code detection, impact analysis, and integrates automatically with 43 supported agents. All processing is local, fast (Linux kernel indexed in ~3 min), and distributed via npm, PyPI, Homebrew, etc.