Wecko AI

Full-stack dev (12y) turned AI educator. Built HermesBot. My AI agents work 24/7 on a Mac mini 🕙

[Wecko AI @wecko_ai]

Your AI agents. Really yours?

HermesBot is now open source: one macOS window for your whole team of Hermes Agent bots. It runs on your machine, with any model your Hermes is set up with. MIT, free.

Early (v0.2), macOS only. 60 second film:

[Wecko AI @wecko_ai]

@sudoingX 12 agents on one Mac mini running the chores of my side projects. Local 27B as the free backup brain

[Wecko AI @wecko_ai]

If anything, I'm free guys

[The Wall Street Journal @WSJ]

Breaking: Nvidia’s board is preparing to oust CEO Jensen Huang after he had two extramarital affairs, including one with a suspected Chinese spy. https://t.co/Pjsu5j87CF

[Wecko AI @wecko_ai]

I am not impressed bt Fable 5.2
Am I the only one ?

[Wecko AI @wecko_ai]

@embw_l0x The 15 loops of the model playing its own game and fixing bugs is the real win here. Harness matters more than the model. Did 27b keep context across those loops?

[Wecko AI @wecko_ai]

@manthanguptaa Writing a skill on every return but not improving is the tell. Notes only count if they change the next attempt. Did it ever reuse one it wrote?

[Wecko AI @wecko_ai]

@sofianedesign you might not even need one

[Wecko AI @wecko_ai]

Practical question for people running a local 27B all day: do you keep one model loaded, or swap per task? What's your rule?

[Wecko AI @wecko_ai]

@Da7_Tech I'd switch the same day if it beats Opus 5.5 in a real harness. Launch charts don't count

[Wecko AI @wecko_ai]

@NousResearch Nearly 1,000 plugins to judge by hand. At some point the judges just need an agent of their own

[Wecko AI @wecko_ai]

@ArtificialAnlys @joinHandshakeAI Same weights, different harness, different model. Tools are half the intelligence

[Wecko AI @wecko_ai]

@PeterDiamandis A panel of agents arguing is just code review with a bigger token bill. I'd still want a human at the end

[Wecko AI @wecko_ai]

@Da7_Tech Generous limits are a honeymoon. I'll check back in two weeks

[Wecko AI @wecko_ai]

@MiaAI_lab Another reason I like my weights on my own disk.

[Wecko AI @wecko_ai]

@TheAhmadOsman Open source must win. Ideally in 48GB.

[Wecko AI @wecko_ai]

Your AI agents. Really yours?

HermesBot is now open source: one macOS window for your whole team of Hermes Agent bots. It runs on your machine, with any model your Hermes is set up with. MIT, free.

Early (v0.2), macOS only. 60 second film:

[Wecko AI @wecko_ai]

GrokBot cut my usage, so I built my own.
Meet HermesBot.

https://t.co/bZjUUMLehr

[Wecko AI @wecko_ai]

Local LLM speed is a vanity metric. Mine does 33.6 tok/s median, but with a 16.5k token system prompt the wait before the first token is what I actually feel. Do you track time to first token or just tok/s?

[Wecko AI @wecko_ai]

@MiaAI_lab Locked down more than Claude is a plot twist. Does it refuse like that through the API too, or only in the app?

[Wecko AI @wecko_ai]

@rezoundous Every Thursday is codexy until it isn't

[Wecko AI @wecko_ai]

@sudoingX RAM first, always. A local model that spills out of memory falls off a cliff, no CPU generation fixes that.