
What is a tiny on-phone AI model that calls tools for you, and why does running AI on your phone matter for privacy?
Tether’s QVAC team released BotPsy-0.8B on November 11 — a 0.8-billion-parameter open-weight model, built on Qwen3.5-0.8B, that runs on the phone itself and specializes in function calling: turning plain requests into calendar entries, messages and settings changes with no cloud round-trip. Why on-device AI matters for privacy, how tool-calling works, and how to check which of your phone’s AI features already run locally.
Short answer: BotPsy-0.8B is an open-weight AI model released on November 11 by QVAC, Tether’s AI research team. It has just 0.8 billion parameters — small enough to run on a phone itself, with the smallest download about half a gigabyte — and it is built on Alibaba’s Qwen3.5-0.8B and specialized for one job: function calling, turning a plain request into the right app action, like adding a calendar event or sending a message, without sending your words to a cloud server first.Why it matters: most AI assistants ship your request to a data center and wait for the answer to come back. An on-device model skips that round trip entirely: it works offline, answers faster, and the request never leaves your phone — which matters when the request involves your calendar, messages or files.
In Tether’s own testing, BotPsy roughly doubled its base model’s tool-use scores (its BFCL v4 score rose from 23.4 to 44.0), though Tether is upfront that it suits low-to-medium difficulty tasks and still makes mistakes — it is a capable errand-runner, not a brain transplant.How tool-calling works, and how to check your own phone:1. What “calling a tool” means: you say “text Maya I’m running ten minutes late,” and instead of just chatting, the model picks the messaging tool, fills in the contact and the message, and sends it — the same pattern works for timers, reminders, files and settings.2. On an iPhone: open Settings > Apple Intelligence & Siri.
Features handled on-device run locally; anything routed further is labelled as using Private Cloud Compute — that label is the tell.3. On Android: search Settings for your assistant (Gemini) and look for on-device options; many quick actions — timers, alarms, summaries of on-screen text — already run on the phone’s built-in small model.4. The airplane-mode test: turn on airplane mode and ask your assistant to set a timer, add a reminder or open an app.
Whatever still works is running on the device itself.5. Audit the permissions: in your privacy settings, check which AI apps can read your messages, files and calendar, and trim anything you do not actually use.Pro tip: on-device is not automatically private — if the app syncs transcripts or backups to an account, copies still leave the phone, so check its sync settings too.
Ask AI about this tip
Send this tip to your favorite AI assistant for a plain-language walkthrough, follow-up questions, or help applying it to your device.