I've been setting up Hermes with a lot of information about me. I gave it a folder of notes about my projects, goals and how I work, then asked it to look through my GitHub folder to see what I'd actually been building.

I wanted it to have more to go on than whatever I happened to mention in one chat. There are projects I've stopped working on, others I've renamed, and things I've changed my mind about. Without that background, it's easy to get advice that sounds reasonable but doesn't fit where I'm at.

Getting the background right

The notes came from information I'd collected across Claude, ChatGPT and Grok. Bringing them together meant going through the parts that disagreed and answering questions about what was still accurate.

Some of the older summaries had treated ideas or suggestions as decisions I'd made. There were also projects whose status needed updating. I had to correct those things rather than just give Hermes a pile of text and assume it now knew everything about me.

Looking through the code was another part of it. I pointed it towards duo and my other repositories so it could compare the notes with what was actually there. A description of what I want to build isn't always the same as what I've finished.

Making it useful day to day

I also gave it the details for my existing Google Docs and Sheets connection and asked it to use that. I wanted it to be able to work with my documents as well as read local project files. Having the background is useful, but I also want help doing the work.

I've been explaining what sort of help I want too. I prefer direct answers and someone challenging my reasoning when it doesn't hold up. I don't need every idea I bring up to be treated as a good one.

Part of what I'm looking for is guidance from something that knows the history: what I've tried, why I stopped, and what I keep getting stuck on. I've talked to it about struggling to stay consistent and switching between projects. I'd like that context to carry into the next conversation instead of having to explain it all again.

I'm still figuring out the setup. Giving it more information doesn't mean every conclusion it draws will be right, so correcting it is part of the process. But this is closer to how I want to use AI: with enough context to help me work through my own situation.