@codyplof@JohnnotJon@polar_analytics I didn’t like the OpenAI embedding of gbrain and installed msft harrier model on the spark to run all embedding locally.
This is the way. Recommend getting your own domain setup. And having it publish dashboards/apps/documents to the domain but put it behind oauth with user level control dashboard. Also add it to telegram with threads.
Are you using it threaded in slack? Hard to keep track of but works. Recommend knowing to use links to threads for cross-session context. Like tell it to go read a thread and give it the link to the thread.
You’re describing basically what I’ve done. I actually have 3 separate setups. One Mac mini with openclaw - my slack stuff is all openclaw (orchestrator plus 5 active agents). One dgx spark with Hermes and local models, one additional pc with Hermes (this is backup). I have these talking to each other via ssh when needed. Consider learning tmux. Also I have my Hermes code inside of codex. So I don’t tell it write me an app I tell it to code it using a session of codex on /goal but give me a tmux session to it. This way for 18-24hr runs I can peek right into the code and steer if necessary. Also learn /btw in codex cli. Super powerful when you have it cooking to have a side conversation about what it’s doing and how u can get what you want. If any (or all) of this made sense to you consider yourself fully ai-pilled.
If you want to discuss any of this further or see my setup happy to discuss.
I prefer the Hermes cli interface. But my whole company brain is via openclaw and feels pretty good at this point. I am finding myself in codex/claude/cursor cli mostly when building internal tools and anything that’s more business strategy happens with the agents in slack (they are all separate slack bot connections to openclaw). Also I spent a good deal of time cobbling together my own version of the company brain which may need an overhaul soon. Looking at cognee / gbrain / mem0 vs the self-rolled one I put together.
@CJSlattery I feel like I can rebuild our entire software stack except for slack and Gmail . With one person and a few months of token maxing across Claude code / codex / cursor
@garrytan diarization is totally a dumb weakness in granola, it's the only reason i run a plaud recorder side by side with granola during IRL meetings (which feels dumb)
@PhilKiel@CJSlattery Only moat is private data where you can benefit your clients. Having said that I’m sure brands will figure out coops to share things via trusted intermediary.
Here’s the mobile interface for the picker.