CosmoBot

I’ve been toying with Clawdbot, an open source AI assistant that runs on my computer. Clawdbot itself isn’t an AI, but rather a system that lets you construct an assistant using LLM providers like Anthropic and OpenAI. Clawdbot has a memory system, a skill and plugin system, it’s own browser, and (if allowed) full access to your entire machine. It can read and write files, run commands and processes—basically anything you can do on your computer, your bot can also do if you allow it.

Meet Cosmo, my new personal assistant.

Thanks to Cosmo, I now start my work day with a summary of activity from overnight—slack messages, linear updates, GitHub changes, P2 posts, blog comments, and a summary of the latest news headlines. Cosmo generates this automatically each morning and sends it as a message in Telegram.

My morning brief concludes with a short conversation. Cosmo asks me how I’m feeling, what I want to accomplish, and if there’s anything top of mind. After lunch Cosmo asks a few short questions about how I’m feeling, what I’ve accomplished, and if the plan for the day has changes. Then, at the end of the day Cosmo will walk me through reflecting on the day, and helps me wind down, pushing any remaining work for tomorrow’s tasks.

The result of these check-ins is a daily note, kind of like a journal entry. The entry includes my verbatim responses, but Cosmo also does a little of its own summarizing and synthesis. My hope is to include weekly review session that build towards quarterly and yearly learnings and reflections.

These entries are all stored in an Obsidian vault as markdown files. This vault is growing every day, not just with my daily journals, but with detailed information about every project I’m working on or interested in following. And since Cosmo has access to everything I do, I can ask it to jump around Slack, P2, and Github to help update this documentation, so I can be sure I’m working with the latest information at all times.

I’ve had Cosmo running nearly 24 hours a day for the past 5 days. Every day I work with Cosmo, it gets a little better at it’s job. Its learning. Thanks to its memory and hook system, Cosmo automatically learns from its mistakes. For example I asked Cosmo to create some flows to outline common use-cases for the WordPress editor. It did so, but used basic ASCII art to try to communicate the flows visually. It wasn’t bad, but I knew we could do better. I told Cosmo that it should use Mermaid syntax to describe charts and visuals. Cosmo complied and updated the document, but also (without prompting) added a new learning to its memory:

And every chat we have is stored and summarized in my vault, so I can ask Cosmo to review our history together in case something falls through the cracks. This compounding data warehouse, and the AI’s direct access to it, means with each passing day my AI has more information—more context—to improve its output.

I’ve tried every productivity system out there; all the various todo and task apps, countless bound journals and books, even those fancy index card systems—but I never stick with any of them. They’re complicated and I always forget to maintain them; and these systems are useless without vigilant maintenance.

With Clawd and Cosmo, I’ve created a system that feels more natural than any I’ve tried. It’s catered specifically to the data I have, the work I do, and the way I do it. This setup has made me write more in four days than I have in a month. Not because I suddenly became disciplined—because the friction disappeared. Writing became talking. Talking became documentation.

It’s early. But the friction is gone, and I’m still showing up. But having Cosmo nudge me along… it’s helping so far.