🧭
the nano family
Eleven apps for people who don't code. Talk to an AI, ask a document a question,
find a lost file, keep a folder safe, play your music — on your own computer,
in plain words, with nothing to install.
Python 3.9+ is the only requirement
works offline
no accounts · no telemetry
800+ automated tests
The apps
Every one starts the same way: download the ZIP, unzip it, double-click run.bat (or ./run.sh).
🧭
nanoHome
One front door for every nano app on this computer — finds them, explains them, starts them.
Download · releases
🧠
nanoLaama
Talk to an AI on your own computer — offline, no account, no key.
Download · releases
📚
nanoDoc
Drop in a document, ask it anything — every answer shows the paragraph it came from.
Download · releases
📊
nanoLearn
Drop a spreadsheet, get an answer machine — which column matters, how well it guessed.
Download · releases
🔊
nanoSay
Have anything read out loud by the voice already inside your computer.
Download · releases
🧲
nanoPick
Find your files by saying what you remember — "a picture called receipt, last month".
Download · releases
🎵
nanoTune
Your music, one page, no account — point it at a folder and press play.
Download · releases
🧰
nanoWrap
The programs already on your computer, with buttons — shrink a video, join PDFs, make a zip.
Download · releases
⌨️
nanoShell
Any program at all, with words instead of flags — its own help becomes a form you fill in.
Download · releases
🗂
nanoGit
Your folder, kept safe and put online, without learning git. Nothing is ever deleted.
Download · releases
🖥
nanoDesk
Every nano-style app you have, one click away, explained like to a stranger.
Download · releases
🃏
nonoForge
Pick a card, answer two questions, press one button — you have a working app.
Download · releases
The engine room
For people who want to see the gears — how a small AI is built, trained and run.
~1,400 lines of dependency-free C: transformer, BPE tokenizer, hand-rolled thread pool, int8 quantization.
A Llama-2 transformer in raw PyTorch — 16.9M parameters, 35 hours of CPU — exported to the engine's format.
Describe an architecture, train it, export it — then teach it to answer with SFT and DPO.
Start here
- Install Python if you don't have it — python.org/downloads (tick Add to PATH).
- Pick an app above, open its Releases page, download the ZIP.
- Unzip it anywhere. Double-click
run.bat (Windows) or run ./run.sh (macOS / Linux).
- A browser window opens with the app. That is all there is.