AI models over BTX.
Running on your computer.

Fetched peer to peer, checked piece by piece, run locally. No account. No cloud.

Download for Mac Beta

v0.8.4 · Apple Silicon · macOS 13 or later · signed and notarized by Apple · updates itself (from 0.7.1, download it once more)

v0.8.4 · Windows 11 · x64 · WSL2 (Windows may ask once to install it) · unsigned: SmartScreen asks once · updates itself (from 0.7.1, download it once more)

Linux · x86_64 · .deb for Ubuntu 24.04+ and Debian 13+ · AppImage for glibc 2.38 or newer

Release notes and SHA-256 · Send feedback

Live since 12 October 2026 · Beta 0.8.4

Experimental indie software, made for fun by easyBTX. Not an official BTX project and not affiliated with the BTX developers. The source code will be published once it has matured.

AInode’s first screen: choose a size, then Start.
01 Choose How much disk it may use. The only setting.
AInode fetching a model from the public host, piece by verified piece.
02 Get Models arrive in pieces. Every piece is checked.
AInode chat with IBM Granite 4.2 3B, answering on the Mac.
03 Chat The model runs on your computer. Your words stay there.
1 public host, live
3 open models: IBM, Qwen, OpenAI
64 s for 2.2 GB, first test
0 accounts needed

Verified

A link names the exact bytes. Anything else is refused.

Private

No account, no cloud. Nothing you type leaves your computer.

Free

Nothing to pay, nothing to earn. Hosts serve for free.

For developers

First public host: 20.196.8.98:29447, BTX v0.34.15, unmodified.

IBM Granite 4.2 3B · 2.2 GB · Apache-2.0 btx://pq4admq4hc7nnvddga9z0flrvg8rhxld390gxdjae3nt8gywty205nn7yhruydsxdkx8k9h87c9uj3z3wcs9h
Qwen3 8B · 5.0 GB · Apache-2.0 btx://pqsltmv73ntyl6x7hf8kclk8fkcxnccsgfy30lmkuqqx0mc5uzhpd9uslhqz2s9x0874vky5yst7nl270yrxm
OpenAI gpt-oss 20B · 12.1 GB · Apache-2.0 btx://pqcjx57rargtrp5us2cgvfdttffzxrfnwfwyrd3m63u9n8wtjqpvs2zf326d025uwrya6wy8tx69mxuxfntv9
IBM Granite 4.0 350M · 237 MB · Apache-2.0 btx://pq33nsv5hxj5mgrs4r4256h80g0y6n7uqhkv0j56cs3nrfq2c9jt686pm507apg2ygtvhd7uc43maj58gc8hu

With BTX’s own helper: -modelpeer=20.196.8.98:29447, then getmodel <link> FREE_ONLY.

Sent upstream

Six bugs found by testing on real links, each sent to BTX with a tested fix. Until #229 ships, AInode uses one host at a time.

  • #228The model network cannot start on a stock Mac.
  • #229A second host for a model breaks its downloads.
  • #230Any upload limit stops a host serving.
  • #231Slow lines throw away finished pieces.
  • #232A model cannot be deleted.
  • #233The macOS source build cannot configure.
Sources and inspiration
Questions
Who makes AInode?
easyBTX, as an indie project, for fun and to see what BTX’s model network can do. It is experimental software: not an official BTX project, not made by or affiliated with the BTX developers. The source code will be published once it has matured a bit more; until then the builds are public on GitHub. Ideas and feedback are welcome there.
Can I download it?
Yes, as a beta: Macs with Apple Silicon on macOS 13 or later (signed and notarized by Apple), Windows 11 on x64, and Linux on x86_64 (both unsigned; Windows warns once). Tell us what breaks.
Which model should I get?
Granite 3B on any computer, fast. Qwen3 8B with 16 GB: the most careful answers, slower. gpt-oss 20B with 24 GB.
Linux and Windows?
Both, since 0.7.1. Linux: a .deb for Ubuntu 24.04 or Debian 13 and newer, or an AppImage. Windows: the engine runs inside a small private Linux that Windows provides (WSL2); the app sets it up on its first Start, and Windows may ask once to install WSL.
Is a model from a stranger safe?
Its bytes are: each piece is checked, so a host cannot swap the file. Whether its answers are right is still yours to judge.
Can I run a host?
Yes, and the network needs more. A small Linux server, about $12 to $33 a month, no BTX needed. The setup scripts will come with the source.