INDIE / MACHINEFree AI That Runs On Your iGPU
Yumon Pet v0.1.0 - Near instant, 100% private desktop AI companion with an aquarium simulation
Designed to be an AI pet that keeps you company when you are at your desk. Yumon runs completely offline on your local integrated graphics (iGPU) using WGPU, meaning zero monthly fees, zero latency, and zero telemetry. Watch living Yumon swim across a living 3D aquarium while Yumon explores daily tech launches and chats with you locally.
Notice: Windows Defender SmartScreen Popups
Because we are an independent indie software team and this is a brand new release, Windows Defender SmartScreen may show a blue prompt stating "Windows protected your PC - Microsoft Defender SmartScreen prevented an unrecognized app from starting".
SmartScreen relies on cryptographic reputation that builds up over time across millions of downloads. As an indie company releasing free software without an expensive corporate enterprise certificate, new version hashes trigger this warning until reputation is established.
- Click More info in the blue SmartScreen window.
- Click the Run anyway button that appears.
- Yumon Portal will start instantly and stay resident on your desktop.
Prefer to verify the build yourself? Yumon Pet is completely open source under the MIT license. You can inspect every line of Rust code in our repository, verify the dependencies, and compile it yourself with a single command:
Near Instant, Zero GPU Burn
Traditional LLMs require multi-gigabyte models and dedicated graphics cards that spin up fans and drain battery. Yumon Pet uses our novel Kilo Talk architecture (4,096 tokens) paired with a CPU N-gram statistical decompressor that executes in microseconds. It runs effortlessly on standard integrated GPUs (Intel UHD 770, Iris Xe, AMD Radeon, Apple Silicon) with minimal memory footprint.
3D Aquarium Simulation
Yumon is not just another boring chat window. As shown in the preview screenshot above, the desktop companion includes a living 3D aquarium simulation built right beside your daily workspace. Watch tropical species - including Neon Tetras, Flame Gobies, Golden Guppies, and Emerald Cories - swim around natural slate rock caves, air stone bubblers, and swaying anubias plants.
Daily Product Hunt Radar
Curious about what the tech world is shipping? Yumon Pet continuously aggregates top daily product releases from Product Hunt and shares autonomous thoughts, observations, and feedback every few minutes. You can also open the chat tab to discuss any launch or bounce your own ideas off Yumon.
100% Private By Design
No account creation, no sign in, no telemetry, and no remote model endpoints. Your messages, questions, and conversations remain on your physical hardware. When you close the app, nothing is leaked to third-party cloud servers.
Open Source (MIT)
We build in the open. The model architectures, tokenizer vocabularies, training pipelines, and native Tao/Wry desktop frontend are published in full on GitHub. Anyone can inspect, fork, or build their own custom pets.
Company At Your Desk
Whether you are working through a long coding sprint, writing documentation, or studying, having a lightweight, animated companion on your secondary monitor makes the workspace feel alive without consuming 50% of your CPU.
YumonPortal-v0.1.0-windows.zip
The release archive contains everything required to run Yumon Portal on Windows out of the box without installing Python, PyTorch, or CUDA toolchains:
- 1.yumon_portal.exe (~25 MB): Standalone Windows executable with embedded Kilo Talk neural weights, BPE tokenizer, and 3D aquarium webview.
- 2.ngram_decompressor.bin (~345 MB): High-accuracy statistical N-gram language decompressor table providing instantaneous microsecond vocabulary expansion.
Audit the Source Code on GitHub
Trust in modern desktop software starts with full source visibility. Every component of Yumon Pet is available for review in the public GitHub repository. You can verify that there is no hidden telemetry, no tracking pixels, and no private data harvesting.