🧠 A private AI on the device you already own β€” free, no cloud

You don’t need a new phone, a powerful PC, or the cloud. A private AI β€” one that answers your questions offline, with no account and no one watching β€” can run on the old device gathering dust in your drawer. Here’s the easiest free way for whatever you own. :brain:

:compass: Start here β€” pick your device

What you have What to do
:laptop: An old laptop or desktop llamafile β€” one file, double-click, chat. No install.
:mobile_phone: An old Android phone PocketPal β€” a free offline chat app, no root needed.
:globe_with_meridians: Just a browser, nothing to install WebLLM β€” a model runs right in the tab.
:strawberry: A Raspberry Pi or cheap board llamafile on Pi β€” the same one-file trick.

:warning: The honest bit first: RAM is the real limit β€” the model plus a little breathing room must fit. Old cheap phones only run small models; a very old iPhone/iPad on old iOS genuinely cannot without a jailbreak. Answers are slow (a few words per second), not desktop-fast. Keep that in mind and the rest is all upside.

:puzzle_piece: What made this possible

smolmaxxing is a free, MIT project that proved a 16B AI brain can run on a 1 GB TV box and a 2013 iPad β€” by patching llama.cpp to keep the busy parts of a model in RAM and read the rest from storage as needed. Every number was measured on the real device.

Device Everyday model Biggest model
:television: TV box (~1 GB, no root) Qwen3.5-2B @ 7.7 tok/s Ling-mini 16B @ 5.1 tok/s
:red_apple: iPad Air 2013 (~700 MB) Bonsai-1.7B @ 24–26 tok/s Ling-mini 16B @ 2.6–4.5 tok/s
:telephone_receiver: Phone (~3 GB free) Qwen3.6-35B @ ~5 tok/s Qwen3-Next 80B @ 2.8 tok/s

Full logs: TV box Β· iPad Β· phone Β· 160 experiments

:hammer_and_wrench: The free toolbox, by device

Device Tools (all free)
:laptop: Old laptop llamafile Β· llama.cpp builds Β· LM Studio
:mobile_phone: Android PocketPal Β· ChatterUI Β· Maid Β· Termux
:red_apple: iPhone/iPad PocketPal Β· Google AI Edge Gallery (recent devices)
:globe_with_meridians: Browser WebLLM Β· transformers.js
:strawberry: Pi / cheap board Pi llamafile guide

:gem_stone: Ungamed rare gems

:ice: NIGHTRUN boots a chat AI straight from a USB stick with no operating system at all.
:mobile_phone: BigMoeOnEdge streams a ~284B model on a 12 GB phone by reading only the parts it needs β€” needs fast storage and ~91 GB free.
:pinching_hand: llamita.cpp runs an 8B 1-bit model on a $35 Pi; project-willamette brings 1-bit AI to old Pentium-class CPUs.
:straight_ruler: Not sure what fits? FitLLM and LocalLM Advisor take your specs and name the biggest model you can run.
:toolbox: GGUF-my-repo turns any model into a small free file right in your browser; awesome-mobile-llm is a catalog of tiny models.

:pinching_hand: The same idea, pushed to the absurd

Project The extreme
PFor 180M params on a $6 ESP32 chip.
ps2-llm A model on the PlayStation 2.
iiama2.c On a Nintendo Wii, 61.9 tok/s.
esp32-ai 28.9M params on an ESP32-S3.
LLM-on-anything The whole hidden menagerie.

:hammer_and_wrench: Make it actually useful

:studio_microphone: Talk to it with whisper.cpp; talk back with Piper or eSpeak-NG. :globe_with_meridians: Serve it to every device at home via llama.cpp web UI, or chat with your own files using sqlite-vec. :house: Keep it on with Node-RED or offline Wikipedia via Kiwix; a 3 MB HermitUI page and a single-file llamafile reach it from any old machine.

:magnifying_glass_tilted_left: Dig deeper β€” the original build notes

:open_book: How it works Β· what didn’t work Β· all model files Β· TV box setup.
:1234: The 1-bit frontier: vlut.cpp (paper) Β· BitNet 1.58-bit (GGUF) Β· Ternary Bonsai.
:white_check_mark: Verified prebuilt Android binaries; the old-iPad route used checkra.in.

:balance_scale: Fine print: Speeds are author-reported and headless; tiny models can ramble. Model files carry their own licences, not MIT. The jailbreak is only needed for the very old iPad experiment β€” every β€œstart here” path above avoids it. Got a small device? Submit your numbers.

Free & open, built by MohammedRashad / Hyperion AI (Nov 2026). :brain: