PIP SUITE

Tiny AI minds that live on your device. Free, open source, and fully offline.
single web pagesno accountsno servers runs on a phoneMIT licensedRWKV-7
🏠 Pip's Room
A home for one small AI resident. They keep a living memory of everything you've shared. They think, wonder and dream while you're quiet, and keep a journal. Ask them to remember something and they can bring back an old conversation word for word. Nothing is hidden: you can see exactly what the model reads, watch their reasoning, rewind a turn, choose their voice, and export everything as readable text.
Open Pip's Room →
💬 Multi-Pip Chat
Put several small models in one room and let them talk to each other, or join in yourself. Each one gets its own voice. Export the whole chat to study it.
Open →
🕸 Junto Orchestrator
A workbench for a swarm of small "Pip" models working together. Load several and see how they divide up a question.
Open →
📚 Dialogue Shelf
Drop in several Multi-Pip Chat exports and see how alike or different the conversations turned out.
Open →
🔍 Will It Run Here?
A quick check of your device: memory, speed and browser features, and an estimate of how fast each model will think on it.
Open →

Getting started

  1. Open an app above. Or download its .html file and open it from your own device: it works fully offline, forever.
  2. Get a model file (below). It's downloaded once and stays on your device.
  3. Load it with the + / Load Model button and say hello.

Every app has a ? help button that explains everything it can do.

Which model?

ModelFileGood for
G1d 0.1B, Q8_0
~210 MB, recommended to start		 rwkv7-g1d-0.1b-Q8_0.gguf Phones and older computers. Charming, quick, a little dreamy. Best with thinking off.
G1d 0.4B, Q6_K
~400 MB		 rwkv7-g1d-0.4b-Q6_K.gguf Steadier conversation, and it can actually reason. Turn on Think to watch it work things out.

These are BlinkDL's RWKV-7 "G1d" models (Apache 2.0), converted to GGUF by shoumenchougou. Official .pth and .safetensors checkpoints load too. Both files are tested in our engine: against the original checkpoints, the 0.1B Q8_0 picks the same next word 99.0% of the time, the 0.4B Q6_K 94.8%. For a 0.1B, prefer Q8_0 over Q4: we measured Q4 costing a model that small noticeably more than it costs a larger one.

Good to know

Private by design. Everything runs in your browser. Nothing you say is sent anywhere, and there are no accounts, trackers or servers. Your resident lives in a file you can export and keep.

A note on what they say. Small models tend to be charming and benign, and they're far too small to help anyone do anything dangerous. But they learned from a huge amount of human writing and only loosely follow what's in front of them, so now and then they say something strange or offensive, especially in dreams and other idle thoughts. There's no filter, on purpose: these apps are for seeing what a small mind really does. User discretion advised.

Not a person, not a replacement for people. Think of it as a terrarium with glass walls: a low-stakes way to experiment with an AI companion and see how one actually works.