Your AI Clients Don't Talk to Each Other. This Project Gives Them a Shared Brain.

You've probably got more than one AI chat client open right now. Maybe Claude Desktop for the heavy lifting, Cursor for code, OpenClaw for messages. Each one is helpful in its own lane, but none of them knows what the others are doing. Your notes live in one, your files in another, and nothing remembers a fact from yesterday's conversation. autoMate is an attempt to fix that by giving all of them a shared layer they can read and write to.

What It Does

autoMate describes itself as "a smart NAS for AI" — a warehouse you own that sits behind whatever chat client you happen to be using. It runs on your machine and stores notes, files, reminders, and cross-session memory in ~/.automate/, backed by SQLite and encrypted with Fernet. You connect your AI client to it over MCP or HTTP, and that client can then write into your notes, drop files, schedule reminders, search your library, and call real-world tools.

The architecture is a hub-and-spoke setup. autoMate is the hub; OpenClaw, Claude Desktop, Cursor, Cline, Kimi, GPT, or your own scripts are the spokes. Once connected, a client gets access to autoMate's tools — things like search.find (BM25 search over notes and files), notes.read, files.list, and audio.transcribe — plus a top-level automate tool that runs autoMate's own agent loop on demand. The key point is that the data is shared: what one client writes today, another can read tomorrow.

You can also skip external clients entirely and use autoMate's built-in web chat, which runs its own agent loop against the same backend and the same data.

Why It's Cool

• The cross-vendor memory problem is real, and this actually addresses it. Most tool-calling setups are scoped to a single vendor's ecosystem. autoMate's pitch is that the memory and files outlive any one client — pick the chat tool you like today, switch tomorrow, and the library comes with you. That's a genuinely useful framing.

• The onboarding is written for AI readers, not just humans. This is the clever bit. After install, you open Settings, click "Copy install text," and get a single markdown blob with your URL and token already filled in, broken into per-client sections. You can hand-edit your config, but you can also paste the blob into another AI and say "set this up for me" — the text is structured so an AI can find the right section and edit the right file. It's a small design decision that says a lot about where the project expects its users to be working.

• Storage is content-addressed and configurable. The file vault deduplicates blobs, and the README notes you can point storage at an external SSD. For something billed as a NAS replacement, letting you choose where the bits actually live matters.

• There are enough entry points that you're not locked into a stack. Python package, standalone binary for Windows/macOS/Linux, Docker for a headless box, a browser extension to drive your real Chrome, and a PWA you add to your phone's home screen. Same backend, different doors.

• The tool surface is broad. Beyond notes and files, it mentions shell, script, browser, desktop control, and 31 SaaS APIs, plus reminders that push to your phone via the extension. Whether you need all of that is another question, but it's there.

How to Try It

Pick whichever install path fits your setup:

# If you have Python
pip install automate-hub

# Or grab a standalone binary from Releases (no Python needed)
# https://github.com/yuruotong1/autoMate/releases/latest

# Or run it in Docker
docker run -p 8765:8765 ghcr.io/yuruotong1/automate:latest

Then just run:

automate

Your browser opens to http://127.0.0.1:8765, and a wizard walks you through picking a model, pasting an API key, and optionally wiring up an AI client. If you want the phone experience, open the hub URL in mobile Chrome or Safari and add it to your home screen — no app store involved.

For client setup, go to Settings, click "Copy install text," and either edit your config by hand or paste the blob into an AI and let it do the work. OpenClaw users paste into the bundle-mcp section. Full details are in docs/channels.md.

The repo is at github.com/yuruotong1/autoMate.

Final Thoughts

autoMate is aimed at people who've already accepted that they'll use multiple AI clients and are tired of each one starting from zero. If that's you — and for a lot of developers it probably is — the shared-memory and file-vault model is worth a look, especially if you want your data sitting on your own disk rather than in someone's cloud. It's a young project with a lot of surface area, so expect to spend some time on setup and testing which tools you actually need. But the core idea, that your AI tools should share a warehouse you own, is one more projects should be taking seriously.