It reads everything.
It only bothers you when it matters.
The Machine reads your email, calendar, chats and phone notifications, finds the few things that actually need you, and sends them to Telegram as short cards. Tap Do it and your own agent carries the task out.
Alex asked about the annual plan on Tuesday and hasn’t had a reply. Their trial ends Friday. I can draft a short answer with the numbers.
A colleague who reads the firehose for you
A small service collects your signals into one private database. Your OpenClaw agent reviews them in the background and only speaks up when something is worth your time.

Reads, never writes
Incremental, read-only collectors for each account. Everything lands in one SQLite file with full-text search.

Finds what needs you
It checks whether something is already handled, really yours, and worth it before it says anything. Saying nothing is normal.

Does the actual work
Drafts the reply, makes the fix, does the research, or asks a colleague and waits for their answer. The result lands on the same card.

Gets better with you
Dismissals, snoozes and the notes you write steer later reviews. Tell it something once and it remembers.
The things that fall through the cracks
Not another notification feed. Each card says what matters and what it can do next, in two or three sentences.
Everywhere work happens
Each account has its own boundary, so work and personal never mix. A new source is one small Python file.
Your data stays on your machine
The service has no model of its own. All reading and thinking happens in the OpenClaw agent you already run, with the model provider you chose. Local models work too.
- One private database. Plain SQLite in your home directory. No cloud, no account, no telemetry.
- Nothing goes out without you. Emails, Slack and LinkedIn messages are sent only after you approve that exact draft. A send that might have failed is checked, never repeated.
- Do it means exactly that. Tapping Do it approves what the card offered, and nothing beyond it.
- Dashboard on your tailnet. It only serves your own Tailscale identity. There’s no public endpoint.
- Quiet when you are. Quiet hours at night, work waits for Monday, and one morning digest for everything that can wait.
Try it on last week first
The preview has your agent review the last seven days once and shows the cards it would have sent. Nothing is sent and nothing is done in your accounts.
Linux with systemd · Python 3.12+ · Node.js 24 or 26 · OpenClaw 2026.10.2+ with its Telegram channel · Tailscale for the dashboard.
# install next to your OpenClaw workspace git clone https://github.com/JeroenCorthout/TheMachine.git \ ~/.openclaw/workspace/tools/personal-assistant cd ~/.openclaw/workspace/tools/personal-assistant ./install.sh # see what it would tell you, without sending anything ./install.sh --preview # happy? turn on reviews and Telegram ./install.sh --activateRead the full guide
Let it read. You get on with your day.
Early release, for people who already run OpenClaw. Feedback and connectors are very welcome.