<lm/prompt>
on-device | 0 bytes sent | tok/s | online | up 00:00
// weather
loading local conditions…
feels
wind
// moon
next full
// sky
sunrise
sunset
daylight
// local
day 0% elapsed
// world
new york
london
tokyo
barcelona
// your machine
webgpu
gpu
shader-f16
memory
cpu
os
browser
display
// network out0 req
to AI servers: none
// location
region
used only to fetch weather
km/h wind
· gusts ·

<lm/prompt>

an intelligence running entirely on your machine. ask it anything — nothing you say leaves this tab.

🔒 your own private AI, running on your machine — nothing you type ever leaves this browser.

// private by design
cloud AI
chatgpt · claude · gemini
→ your words leave your device
lm/prompt
your own model · your own hardware
✓ your words never leave your device
no account, no privacy policy to trust, no setting to toggle — your words have nowhere to go but your own machine. that is the one thing cloud AI fundamentally cannot offer.
works on any GPU — NVIDIA, AMD, Intel, or Apple — or no GPU at all. no install, no special hardware.
talk to the singularity
a model running entirely on your machine — here, in this browser tab. no server, no internet, nothing you say leaves your device.
click to wake the singularity
click the box to wake the on-device AI~764MB model loads onto your GPU, cached after first load · runs 100% in your browser
enter twice to submit
strict input filtering: slashes (/ \), percent (%), and hidden/zero-width characters are removed automatically.
tech / diagnostics
fun stuff — let the local AI guess things about you
runs on your GPU · playful guesses, not facts

// contact tomATcanadacompDOTca
_ a small mind, running where nothing can watch it.
debug log
live events · model load, GPU, network, errors
// generate a document · .docxon-device
describe a doc or form — the on-device model drafts it and your browser builds a real Word file. nothing leaves this tab.
// news
loading headlines…
// engine
statusidle
first token
tokens (session)0
context0 / 4096
est. vram
est. footprint · vram not directly readable
// privacy
your text sent0 bytes
AI server calls0
accountnone
cost$0.00
cookies0
offlineready
// session
tokens0
rewrites0
local runs0
uptime00:00
history0 saved
// tools
rewriterlive
reformattersoon
rephrasersoon
taggersoon
summarizersoon
explainersoon
// keys
submit↵ ↵
clearesc
// ai says
your model runs here on your machine. paste a rough prompt in the box and it gets cleaned up — nothing leaves this tab.
// how it runs on your machine
tier 0
chrome nano
built-in model when chrome has it. no download.
tier 1
webgpu · gemma
small model on your GPU — any brand.
tier 2
instant fallback
no GPU? rule-based, any browser.
// activity
// /commands
rewrite a prompt/fix
change the tone/tone
summarize text/sum
// achievements
first wakeoffline run7-day streakall models
// stack
webgputransformers.jschrome prompt apivanilla jsstatic html
single static page · no backend for the tool
lm/prompt:~$ console v0.1.0 · built 2026-06-24 14:47 EDT · awaiting input