Reverse-engineered enough of the Divoom MiniToo's Bluetooth display protocol to be able to stream a tiny pixel avatar to embody my Hermes agent. T-pose proof of concept @NousResearch
@alxfazio@flydotio@Zai_org Why not the agents SDK? I have a similar setup and found that --print chewed through tokens, I think it loads the entire CC system prompt each time. A neat pattern I've found is to use the discord channel history as sort of a parallel context window for longer term memory
@alexhillman Yes! The best part is telling it to modify its discord outputs from within discord itself. Asking for a new slash command, and then being able to use it seconds later without ever leaving the channel is magical
@irl_danB Sounds very familiar, especially these days. For what it's worth I think OpenProse is a brilliant idea; the 'fuzzy vs procedural' boundary is the most interesting space to explore with LLMs imo
@ejc3@adocomplete https://t.co/Fjs9nni0EX I'm building this. You can use Claude Code or any other CLI tool. Lmk if you'd like access to test it out (it's free and self-hosted)
@jonas I've been working on this for myself. Planning to open source it soon, but let me know if you want access to try it out in the meantime. It uses worktrees for separating workspaces within the same container while allowing agents to easily communicate
@jonas https://t.co/pceh5xojwl it's kind of a batteries-included version of this, with a bunch of UI helpers for input and source control. But the server-client architecture is also nice for things like notifications and webhooks
@jonas I've been working on this for myself. Planning to open source it soon, but let me know if you want access to try it out in the meantime. It uses worktrees for separating workspaces within the same container while allowing agents to easily communicate