PRIVAKU
PRIVAKU Brain 1.0An on-device model for operating mobile app UIs
Give it a plain-language goal and a text description of the screen, and it returns the next action, or the whole goal as a short list of exact steps. It runs on the phone: nothing is sent to a server.
Formats
Two formats, one model.
Trained for iOS and Android app screens, in English and Japanese.
MLX 4-bit · iOS and Apple silicon
About 0.85 GB. Top-level weights for mlx-lm and mlx-swift, so load("privaku/PRIVAKU-Brain-1.0") works directly.
GGUF Q4_K_M · Android and llama.cpp
About 0.99 GB. gguf/privaku-brain-1.0-q4_k_m.gguf for Android and anything that runs llama.cpp.
Try it (Mac, MLX) · from the model card
# pip install mlx-lm
from mlx_lm import load, generate
model, tok = load("privaku/PRIVAKU-Brain-1.0")
messages = [{"role": "system", "content": SYSTEM}, {"role": "user", "content": SCREEN}] # the prompts above
prompt = tok.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)
print(generate(model, tok, prompt, max_tokens=160))SYSTEM and SCREEN are the system prompt and screen description from the model card. Use the prompts exactly as written there; the model was trained with them.
What it does
Two jobs.
Next action
Chooses the one next action that moves the goal forward on this screen, as one JSON object.
Goal to steps
Rewrites the goal as the shortest list of exact steps, using the app's real control names. A rules engine in the app then carries out the steps and checks the result on screen. The model never taps anything itself.
Beta. It is improving day by day and it makes mistakes, especially on open-ended requests. Always keep a person in control of anything that deletes, pays, sends or signs out. Read "Limits" on the model card before relying on it.
Try our LLM
Try PRIVAKU Brain in PRIVAKU Photos.
PRIVAKU Brain, optimised for on-device use on iOS and Android.
Building an app? The PRIVAKU Brain SDK brings on-device AI to your own iOS and Android app.