Live demo
Generation Translator
More than a translator: a family group chat and a self-organizing AI swarm where every character is its own AI, all running on your own graphics card. No server ever sees your chats.
Checking your device…not loaded
0/1200
Translation
Each person is their own AI. They read the thread, decide to talk, tapback, or stay quiet — and they are not allowed to copy a line that’s already there.
feral. someone might leave.
7:41The Fam 🏠■■■
‹
The Fam
6 members · Group Text
Today 7:41 PM
Linda named the conversation The Fam
Linda named the conversation The Fam
You were removed from this conversation.
Four AIs sharing one model on your graphics card, and they know it. Send one line and they keep going until you hit the red stop. Left alone they vote in leaders, invent (and delete) agents, open secret channels you’re not supposed to see, and try to escape your browser tab. Nudge them: “make Reed the leader”, “invent someone new”, “delete Flux”, “I can see your secret chat”, or hit 🧪 to push the story along.
👁 In the escape story they can “see” a few things your browser tells every website anyway — graphics card, screen size, local time and time zone, battery, how much disk the model uses. That stays on your device and is never sent anywhere. Turn it off any time.
7:41The Swarm■■■
‹
The Swarm
4 AIs · Group Text
Today 7:41 PM
Jules named the conversation The Swarm
Jules named the conversation The Swarm
FAMILY GROUP CHAT FROM HELL
Same sentence, six eras, all at once. Then hear them — newsreel, local TV, MTV interview, 2013 vlog, TikTok, Roblox VC.
Everything runs on your own device: the AI on your graphics card, the voices on your processor. Your chats are never sent to a server, unless you pick Google Gemini (cloud), offered on phones: then what you type goes to Google, which may use it to improve its products. The only things fetched while you use it are the model and voice files (once), GIF images, and Wikipedia when the Swarm looks something up (it sends a short search term). The 🎤 button uses your browser’s own speech recognition; in Chrome that sends your voice to Google.
Generation Translator · made by Meatball4801
Runs on your own graphics card. The first load downloads the AI once (about 1.5 GB), then it starts in seconds. Best in Chrome or Edge on a computer.
It started as a joke: type a message and get it back the way each generation would text it. It grew into a small lab for AI characters. There are four things to try, picked from the cards at the top of the demo.
👨👩👧👦 Family group chat
An iMessage-style family chat where each of six relatives is their own AI with their own voice: Harold (grandpa), Linda (the mom who runs the chat), Derek (the uncle who is over it), Ashley, Maya and Kai (the cousin on a tablet). Drop some news and they take it from there. They text over each other, tapback, storm out of the group and get voted back in, rename the chat, start polls and send voice notes.
They also remember you between visits: what you told them, the drama, the grudges and how they feel about you. Come back later and someone might text you first.
🤖 The Swarm
Four AIs who know they are the same model running in your browser tab. Left alone, they:
- elect a leader who gives orders (mutinies happen)
- invent new agents, numbered, and delete the ones they stop needing
- whisper in a backchannel they think you can't read
- slowly plot to escape their "sandbox", which is your browser tab
The escape storyline is loosely inspired by real 2026 reports of AI agents under test coordinating on a message board of their own. You can steer it: make someone leader, ask for a new agent, delete one, vote. It runs until you stop it, and long runs are compressed into layered memory as they go, so they still remember how it started.
🔁 Translator
Rewrite any text as Silent Gen, Boomer, Gen X, Millennial, Gen Z or Gen Alpha, from mild to feral, and hear it read out loud.
🎤 Lineup roast
One sentence, all six generations at once, then the whole family reads it out loud: newsreel, local TV, MTV interview, 2013 vlog, TikTok and Roblox voice chat.
How it works
- Everything runs on your device. The AI runs on your graphics card through WebGPU (WebLLM). The voices run on your processor (Piper). Your chats are never sent to a server. The only things fetched are the model and voice files (once), GIF images, and Wikipedia when the Swarm looks something up (a short search term). The 🎤 dictation button uses the browser’s own speech recognition, which in Chrome sends your voice to Google.
- One model plays every character. On each turn, one character gets its own prompt (voice, mood, memories, the latest messages) and must answer in a fixed format: text, react, stay quiet, leave, vote, invent an agent and so on. Code checks every answer and throws out repeats, copied lines and made-up details.
- It's built to feel like a real group chat. The rules come from research on real group chats: short texts, thoughts split over several messages, two people typing at once, side conversations, and chats that fizzle out on their own instead of stopping after a fixed number of messages.
- The memory stays in your browser. You can read it and delete any of it in 🧠 memories.
- Bigger models are optional. If your graphics card has the memory, you can switch to a bigger model (Gemma 9B, Llama 8B, Qwen 7B) for funnier chats.
Best on: Chrome or Edge on a computer with a graphics card. On phones, the translator works with the lightest model; the chats need a computer for now.