On the call tonight, some people were talking about business models, mission statements, directions, agent identities, and so on. One thing I find really helpful: start a Claude project or at least a Claude chat. Give it some basic information about who you are, what you're trying to build, a grand idea, or even a tiny one. Upload any docs you already have. Then tell Claude to interview you about it. You can use that chat or project to generate things like mission statements, and to craft better prompts for your agents down the road. I'd use Whisper or Mac dictation and just speak your answers. Way faster than typing and your thoughts "are not constrained by your typing." Make sure you save the interview output somewhere: Notion, Google Docs, a markdown file, whatever works. This will potentially save you, your agents, and your wallet in the long run. And it forces you to actually think through what you're trying to accomplish. For the interview itself, I'd use a high-capability model. It'll do a better job synthesizing high-level concepts, surfacing sticking points and challenges, and generating a solid to-do list. If it starts going down the wrong track with its questions, just say "that doesn't matter right now" and redirect it. That's my tip of the evening. Have a good one.

Wanted to know which version mac mini should I invest in? Should I get a new one or grab something used on FB marketplace?

We discussed lossless claw on one of the calls recently. A heads up. I installed it and it never wrote to the database. I determined it was because the default was set to 75% of the context window full. If you're using a 1 million token window, it does not save to the database until you've done a whole lot of work. To me that missed the point. I want it to back up regularly, so that I can clear the token window regularly, so that I'm not burning tokens. All that to say, I asked my agent to set the threshold to .05 which is 50 K tokens in the window. It didn't trigger the first time I reached that, because something had control over the compression setting trigger. I figured that out by asking my agent to see if the database was full yet. It had not written anything and then it figured out why. Hope this helps somebody..

Does anyone know if there is a known best OCR for reading messy documents to pair with OpenClaw? When doing research into real estate deals, it involves searching old titles/deed and alot of them are scanned documents on government websites and they can be old sloppy hand written pages.

Good Morning everyone - anyone else using the OpenAI Pro plan with OpenClaw having issues with API rate limit responses today? I woke up to all my bots returning rate limit errors this morning and of course I’m out of town so I don’t have direct access to my host