# 🧠 A private AI on the device you already own — free, no cloud

**URL:** <https://onehack.st/t/a-private-ai-on-the-device-you-already-own-free-no-cloud/327287>\
**Category:** Tools & Scripts\
**Tags:** tools, android, ai\
**Created:** [November 16, 2026, 6:01pm UTC](https://onehack.st/t/a-private-ai-on-the-device-you-already-own-free-no-cloud/327287 "2026-11-16T18:01:57Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Margaret](https://onehack.st/user_avatar/onehack.st/margaret/32/174872_2.png) [@Margaret](https://onehack.st/u/Margaret)\
**Post date:** [November 16, 2026, 6:01pm UTC](https://onehack.st/t/a-private-ai-on-the-device-you-already-own-free-no-cloud/327287/1 "2026-11-16T18:01:58Z")

</div>

![smolmaxxing — a private AI on the device you already own](https://onehack.st/uploads/default/original/3X/7/7/77197c95feb654eb89f642620dd835000ccfcbd2.png)

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

### 🧭 Start here — pick your device

| What you have | What to do |
| --- | --- |
| 💻 An old laptop or desktop | [llamafile](https://github.com/mozilla-ai/llamafile) — one file, double-click, chat. No install. |
| 📱 An old Android phone | [PocketPal](https://www.pocketpal.dev/) — a free offline chat app, no root needed. |
| 🌐 Just a browser, nothing to install | [WebLLM](https://webllm.io/) — a model runs right in the tab. |
| 🍓 A Raspberry Pi or cheap board | [llamafile on Pi](https://projects.raspberrypi.org/en/projects/llm-rpi) — the same one-file trick. |

⚠ **The honest bit first:** RAM is the real limit — 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.

### 🧩 What made this possible

[smolmaxxing](https://github.com/hyperionaidev/smolmaxxing) is a free, MIT project that proved a 16B AI brain can run on a 1 GB TV box and a 2013 iPad — 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 |
| --- | --- | --- |
| 📺 TV box (~1 GB, no root) | Qwen3.5-2B @ 7.7 tok/s | Ling-mini 16B @ 5.1 tok/s |
| 🍎 iPad Air 2013 (~700 MB) | Bonsai-1.7B @ 24–26 tok/s | Ling-mini 16B @ 2.6–4.5 tok/s |
| 📞 Phone (~3 GB free) | Qwen3.6-35B @ ~5 tok/s | Qwen3-Next 80B @ 2.8 tok/s |

Full logs: [TV box](https://github.com/hyperionaidev/smolmaxxing/blob/main/docs/tv-box.md) · [iPad](https://github.com/hyperionaidev/smolmaxxing/blob/main/ipad-air/README.md) · [phone](https://github.com/hyperionaidev/smolmaxxing/blob/main/docs/phone.md) · [160 experiments](https://github.com/hyperionaidev/smolmaxxing)

### 🛠 The free toolbox, by device

| Device | Tools (all free) |
| --- | --- |
| 💻 Old laptop | [llamafile](https://github.com/mozilla-ai/llamafile) · [llama.cpp builds](https://github.com/ggml-org/llama.cpp/releases) · [LM Studio](https://lmstudio.ai) |
| 📱 Android | [PocketPal](https://www.pocketpal.dev/) · [ChatterUI](https://github.com/Vali-98/ChatterUI) · [Maid](https://f-droid.org/packages/com.danemadsen.maid/) · [Termux](https://github.com/ggml-org/llama.cpp/blob/HEAD/docs/android.md) |
| 🍎 iPhone/iPad | [PocketPal](https://apps.apple.com/us/app/pocketpal-ai/id6502579498) · [Google AI Edge Gallery](https://gemma.ai/deployment/ai-edge-gallery/) (recent devices) |
| 🌐 Browser | [WebLLM](https://webllm.io/) · [transformers.js](https://huggingface.co/docs/transformers.js/guides/webgpu) |
| 🍓 Pi / cheap board | [Pi llamafile guide](https://projects.raspberrypi.org/en/projects/llm-rpi) |

### 💎 Ungamed rare gems

🧊 [NIGHTRUN](https://github.com/hardrave/NIGHTRUN) boots a chat AI straight from a USB stick with no operating system at all.  
📱 [BigMoeOnEdge](https://github.com/Helldez/BigMoeOnEdge) streams a ~284B model on a 12 GB phone by reading only the parts it needs — needs fast storage and ~91 GB free.  
🤏 [llamita.cpp](https://github.com/coverblew/llamita.cpp) runs an 8B 1-bit model on a $35 Pi; [project-willamette](https://github.com/nangman-infra/project-willamette) brings 1-bit AI to old Pentium-class CPUs.  
📏 Not sure what fits? [FitLLM](https://fitllm.run) and [LocalLM Advisor](https://localllmadvisor.com) take your specs and name the biggest model you can run.  
🧰 [GGUF-my-repo](https://huggingface.co/spaces/ggml-org/gguf-my-repo) turns any model into a small free file right in your browser; [awesome-mobile-llm](https://github.com/stevelaskaridis/awesome-mobile-llm) is a catalog of tiny models.

### 🤏 The same idea, pushed to the absurd

| Project | The extreme |
| --- | --- |
| [PFor](https://github.com/cyfrit/p-for-llm) | 180M params on a **$6 ESP32** chip. |
| [ps2-llm](https://github.com/xaskasdf/ps2-llm) | A model on the **PlayStation 2**. |
| [iiama2.c](https://github.com/brysontang/iiama2.c) | On a **Nintendo Wii** , 61.9 tok/s. |
| [esp32-ai](https://github.com/slvDev/esp32-ai) | 28.9M params on an **ESP32-S3**. |
| [LLM-on-anything](https://github.com/emrehangorgec/awesome-llm-on-anything) | The whole hidden menagerie. |

### 🛠 Make it actually useful

🎙 Talk to it with [whisper.cpp](https://github.com/ggml-org/whisper.cpp/blob/master/examples/stream/README.md); talk back with [Piper](https://github.com/OHF-Voice/piper1-gpl) or [eSpeak-NG](https://github.com/espeak-ng/espeak-ng). 🌐 Serve it to every device at home via [llama.cpp web UI](https://llama.app/docs/webui), or chat with your own files using [sqlite-vec](https://github.com/asg017/sqlite-vec). 🏠 Keep it on with [Node-RED](https://nodered.org/docs/getting-started/raspberrypi) or offline Wikipedia via [Kiwix](https://get.kiwix.org/en/solutions/applications/kiwix-server/); a 3 MB [HermitUI](https://github.com/moooff/HermitUI) page and a single-file [llamafile](https://docs.mozilla.ai/llamafile/getting-started/pre-built-llamafiles) reach it from any old machine.

### 🔍 Dig deeper — the original build notes

📖 [How it works](https://github.com/hyperionaidev/smolmaxxing/blob/main/docs/how-it-works.md) · [what didn’t work](https://github.com/hyperionaidev/smolmaxxing/blob/main/docs/lessons.md) · [all model files](https://github.com/hyperionaidev/smolmaxxing/blob/main/docs/models.md) · [TV box setup](https://github.com/hyperionaidev/smolmaxxing/blob/main/docs/tv-box-setup.md).  
🔢 The 1-bit frontier: [vlut.cpp](https://github.com/OpenBitSys/vlut.cpp) ([paper](https://arxiv.org/abs/2512.06443)) · [BitNet 1.58-bit](https://github.com/microsoft/bitnet) ([GGUF](https://huggingface.co/microsoft/bitnet-b1.58-2B-4T-gguf)) · [Ternary Bonsai](https://huggingface.co/prism-ml/Ternary-Bonsai-1.7B-gguf).  
✅ Verified prebuilt [Android binaries](https://github.com/uno-km/amv-llamacpp); the old-iPad route used [checkra.in](https://checkra.in/).

⚖ **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 — every “start here” path above avoids it. Got a small device? [Submit your numbers](https://github.com/hyperionaidev/smolmaxxing).

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