Sno Station
Sno Station is your agents' workstation: open-source software that turns the AI agents you already run into one squad on your own machine. Two heads are better than one, and they get better with every
A free browser extension that hides spam, scam links and abuse in YouTube, Twitch and Kick chat. Runs on your own computer. Early version.
PluginsTypeScript · PythonGitHub @daradudaiOnlineOwner unconfirmed
Have you tried it? One tap is enough. No sign-in needed.
Owners of public repositories under a personal GitHub account can sign in to verify ownership and manage the introduction themselves. No administrator approval is needed.
Sign in, generate a verification file, and commit it to this repository’s default branch within 30 minutes. Files on other branches or forks do not count. Then check the file below to manage the project without administrator approval. Keep the file to retain your access.
If needed, you will go to sign-in. After signing in, return here and press your chosen verification button again.
Run /nomorevibe verify in your project folder to verify ownership and manage the introduction yourself.
Verify in one minuteAsk about other verification options
Live chat without the junk: on-device AI hides spam, scams and abuse in YouTube, Twitch and Kick chat.
Taken from the repository README
Loading the full README. The saved excerpt is shown below.
Live chat without the junk.
A free browser extension that hides spam, scam links and abuse in YouTube, Twitch and Kick chat.
A small AI model runs on your own computer. The chat never leaves it.
Early version · testing now · tiho.chat
Tiho on a live Twitch chat ( MP4 ). Chatter names are swapped for made-up ones.Read the full README ↗2 hidden · show line.KEKW at once, you see it once: KEKW ×37.It's early and it still makes mistakes: some abuse gets through, and now and then a normal message is hidden (one click brings it back). It's checked properly on English and about a dozen other languages; others may work but aren't tested. On older computers it can fall behind in very busy chats.
Needs Chrome or Edge 124+ on desktop.
Add Tiho to Chrome. In Edge, open the same link and allow extensions from other stores.
The ways below install the same extension from a GitHub release, for people who'd rather not use the store.
If you use Claude Code, Codex, Cursor or a similar agent, paste this:
Install the Tiho browser extension for me. Download the latest
tiho-*-chrome.zipfrom https://github.com/daradudai/tiho/releases, unzip it into a folder I can keep, then tell me exactly how to load that folder in Chrome or Edge (chrome://extensionsoredge://extensions→ Developer mode → Load unpacked).
The last click (Load unpacked) is yours: browsers don't let programs install extensions on their own.
tiho-*-chrome.zip from Releases and unzip it.chrome://extensions (or edge://extensions), turn on Developer mode, click Load unpacked and pick the unzipped folder (the one with manifest.json).The release zip already contains the AI model (about 200 MB), so it works offline from the first message.
You need Node.js 22. The model isn't in git, so take it from a release zip:
git clone https://github.com/daradudai/tiho.git
cd tiho/extension
npm ci
cp -r path/to/unzipped-release/models public/ # PowerShell: Copy-Item -Recurse path\to\unzipped-release\models public\
npm run build # then load extension/.output/chrome-mv3
Without the model the extension still runs, but only its simple rules filter. More build options: extension/README.md.
For people with a strong graphics card: a small local server that runs a larger AI model (Cloudflare Clef-Flash, 9B) and lets you write filters in plain words, like "hide backseating" or "hide Elden Ring spoilers". It runs only on your computer and is off until you turn it on.
cd engine
./start.sh # Windows: .\start.ps1
Then turn it on in Side panel → Rules → Custom AI filters. Details and GPU requirements: engine/README.md.
A content script watches only the chat list. Fast rules handle the obvious cases (your rules, floods, repeats, waves) in microseconds. The rest goes in small batches to one shared AI model in the background: Tiho Lite, a multilingual mmBERT model distilled down to 6 layers for the GPU and 3 for the CPU. Each message gets seven scores (toxic, spam, scam, noise, question, positive, negative), and your settings decide what to show, fold or hide.
How the model was made: we collected about a million real chat messages from Twitch, Kick and YouTube, labeled them with a large teacher model and with AI annotators, and distilled the result into a model small enough to run in a browser. On our 2,000-message test set it beats the 9B teacher it learned from (macro AP 0.846 vs 0.721).
More: architecture · model card · training pipeline · benchmarks · privacy
Bug reports and "it hid this wrongly" reports are very welcome; open an issue. For code, see CONTRIBUTING.md.
FSL-1.1-MIT (Functional Source License). You can use, change and share Tiho for anything except building a competing commercial product or service from it, and every version becomes plain MIT two years after its release. The code and the Tiho Lite model weights are under this license. Third-party parts keep their own licenses: see THIRD_PARTY_NOTICES.md.
Built on mmBERT, Clef-Flash, llama.cpp, ONNX Runtime Web, Transformers.js, WXT and Preact, plus the public datasets credited in the model card.