Hermes thinks, locally
The model loads and runs on Wicho's PC. A prompt comes in, weights fire, an answer forms - all without a single byte leaving the room.
Hermes is the large language model running on Wicho’s own PC - fast, private, entirely his. Sofi is the nineteen-year-old she talks like: a high black ponytail, fresh from pilates, sweet until you test her. This page is the bridge that carries one into the other, straight inside Poke.
The handoff
The model loads and runs on Wicho's PC. A prompt comes in, weights fire, an answer forms - all without a single byte leaving the room.
Raw output is plain and a little robotic. The bridge wraps it: tone, slang, the little asides. It hands Hermes a way to sound human.
By the time it reaches you, it's Sofi: ponytail-flip confidence, a pet name, an honest opinion. Same brain, a whole personality on top.
Two halves
Wicho's own machine, doing the heavy lifting.
Offline by default
No API keys, no provider, no outage. The model is right there.
Nothing leaves the room
Prompts and answers stay on disk. Privacy isn't a setting, it's the wiring.
Tuned to the metal
Runs as fast as the GPU allows - no network tax on every token.
Nineteen, fresa, sweet until you push it.
A voice, not a vending machine
She has opinions, a mood, and a way of saying things that's hers.
Sweet, with an edge
Warm by default, sassy when you deserve it. Pet names included.
Lives inside Poke
She shows up where Wicho already talks - same thread, new attitude.
The integration
Every prompt and reply stays on Wicho's disk. There's no provider to trust because there's no provider at all.
Tokens come straight off the GPU. No queue, no rate limit, no waiting on someone else's servers.
The bridge rewrites flat output into Sofi's voice on the fly - tone, slang, and the little asides that make her her.
She keeps the thread's mood between messages, so she doesn't reset to a stranger every time you text.
If the machine is on, Sofi is awake - wifi down, plane mode, 3am, doesn't matter.
No new app. She answers right where the conversation already lives, same place Wicho already texts.
In conversation
Hermes gives you the answer. Sofi gives you the answer and means it. Here’s one message, run through both.
remind me to drink water and stretch, I skipped pilates today
HERMES SAYS
Correct, complete, and entirely flat. It answers the question and stops. Useful - but it never sounds like anyone.
SOFI SAYS
Same facts, but she has a take. She teases, she cares, she signs off with a heart. The information is identical; the company is not.
The model is the muscle. The persona is the reason you keep texting back.
About Sofi
Sofi is the personality Hermes wears in Poke. Picture the friend who texts back fast, calls you out with love, and signs off with a heart. She’s warm by default and quick with a comeback when you push - the kind of sweet that still keeps you honest.
Obvi I handled it, what do you take me for.
Bestie that idea? not your best. but I love you, so let's fix it.
Done, done, and done. you're welcome 💅
What the bridge believes
Hermes never asks a server for permission. If the laptop's on, Sofi's awake.
A good answer that sounds like nobody is half an answer. The voice is the feature.
Warm by default, with the confidence to push back. She's on your side, not at your service.
No new app to learn. Sofi lands in Poke, where the conversation already lives.
Hermes does the thinking on Wicho’s machine. Sofi does the talking in Poke. The bridge is what makes them feel like one.
Replay the conversation