Yan

Software Engineer. I live to code useless stuff.

[Yan @vikmeowin]

@GonzaloRTarifa @Angaisb_ Awesome! thanks!
can you share a prompt that you use to build it?

[Yan @vikmeowin]

@GonzaloRTarifa @Angaisb_ i mean - your Minecraft clone

[Yan @vikmeowin]

@victormustar Hi, Victor, I just sent you a question in a dm about one of your projects. I'd really appreciate it if you could find the time to reply.

[Yan @vikmeowin]

@victormustar Hey Victor, I'd really like to talk to you via dm. Please accept my message request.

[Yan @vikmeowin]

@victormustar Hey Victor, I've sent you dm, could you check it please? Thanks!

[Yan @vikmeowin]

@BRICSinfo It's NOT a trap!

[Skyler Miao @SkylerMiao7]

This is an outstanding project. I really like it; the taste is excellent.
M2.7 has good performance on it. We are also considering further iterations and optimizations.
@NousResearch

[Jorge Lindo @jorgexlindo]

Thank you @NousResearch 🩸🔬 Our cancer & blood biobank running now with #Hermes

[Yan @vikmeowin]

@itsmetamike @NousResearch 🔥🔥

[metamike @itsmetamike]

Last week I pivoted from OpenClaw to Hermes Agent because A) I love @NousResearch as a brand and B) it was gaining some traction.

I built a self-improving agentic system that:

[witcheer @witcheer]

my @NousResearch <> @stackwalnuts setup that fixed most of the memory issues I was hitting

most AI agents have a memory problem. they stuff everything into one file: MEMORY.md, SOUL.md.
that gets loaded on every single interaction. it bloats. it gets noisy. the agent starts hallucinating from its own context. the more it remembers, the worse it gets.

@stackwalnuts fixed it without touching Hermes's own memory system.

ALIVE organises context into "walnuts", structured md files split across five domains. separate from MEMORY.md. separate from SOUL.md. loaded only when relevant, not on every message.

I set up 5: witcheer (personal brand), yari-finance (day job, CDP protocol), arcana (redacted), oz-agent (the agent itself), micro-entreprise (invoicing).

Oz gets persistent, growing memory across all my projects without bloating the core files Hermes reads on every interaction.

----

but out of the box, Hermes couldn't see walnut files. so I built the bridge myself, custom Hermes skills that load walnuts into the current session on demand.

type "walnuts" or "status" on telegram and Oz reads all 5 files from disk, synthesizes a cross-venture view.

that insight was impossible before. Oz couldn't see across projects. now it can.

then I wired the crons.

~/ crons WRITE to walnuts.
created https://t.co/6XvAtCfmQm - prepends log entries to walnut files after every research session. every session leaves a trace.

~/ crons READ from walnuts.
injected walnut reads into the PRE-READ section of 6 cron prompts. before researching, each cron now reads the relevant walnut's tasks, insights, and current state.

I also run Claude Code for planning and complex work. both systems read and write to the same ALIVE walnuts on disk. two AI systems, one context layer. md files that both understand.

update walnut tasks → crons read them → research aligns to priorities → findings update the walnuts → adjust tasks → cycle repeats.

Hermes agent's biggest weakness was memory. @stackwalnuts fixed it.

[Yan @vikmeowin]

@ivanfioravanti @NousResearch Fr fr

[dotta 📎 @dotta]

Learn how Cathryn uses Paperclip to manage her team of agents 📎

[Yan @vikmeowin]

@somewheresy People underestimate the importance of this

[Yan @vikmeowin]

@theyfwzik It's good, try it

[Yan @vikmeowin]

@ehizogietweets Exactly my thoughts fr

[Yan @vikmeowin]

@sakda_blackcat Crazy how AI nailed the mechanics