I’ve been dying for something like this.
Clawdbot, for anyone who hasn’t run into it yet, is Peter Steinberger’s open-source assistant, first published in November. It runs on your own machine, and you talk to it through the chat apps you already have. Its own pitch is blunt — it “clears your inbox, sends emails, manages your calendar, checks you in for flights,” all from WhatsApp or Telegram — and its author is candid that it isn’t finished and has sharp edges.
The idea I’d chase next: have it spin up a revised version of itself, partitioned, so that one machine can effectively give an account to every member of a family. An assistant for you and another for your spouse, each walled off from the other. Partitioned matters. A family sharing one assistant is a family sharing one inbox, and nobody wants that. Separate memories, separate permissions, one box under the desk. I’d be working on that fork next if the idea of adding skills and a smart Puppeteer weren’t occupying my time — I’m building those into Ragu’s enterprise version of Clawdbot.
Meanwhile I’m wondering about installing it on my primary personal machines. My thinking is that it carries the same risk profile as what I’m already doing with Claude Code, Anthropic’s agent that “reads your codebase, edits files, runs commands” — and the uptick for me is skills, and Telegram as an interface. What makes an assistant dangerous is what it can reach, and mine can already reach the same things from the terminal. The phone just shortens the walk. And Telegram means the assistant is wherever my phone is, which is everywhere, dinner included. Just wondering.
Why I want it, in one small story. Last night I was on my way to dinner and wanted to spin up some work on an idea. I had to wait until this morning. Both systems could have had me 14 hours further along on the build. My workarounds have been less than ideal.
