Your Data, Your Rules
All AI classification runs locally on your device using open-source models. No cloud APIs, no data uploads, no exceptions. Download models once, then work fully offline forever.
100% Local AI
One-time model download (~2GB)
On first launch, Talyx downloads open-source models from HuggingFace: Qwen2.5 for text, SmolVLM for vision, Florence-2 for OCR, and GLiNER for entity recognition. This is the only time Talyx needs an internet connection for AI.
Everything runs on your machine
All classification — documents, images, and videos — runs locally via node-llama-cpp and ONNX Runtime. No API keys, no cloud accounts, no data uploads. Your files never leave your device.
Multi-model AI pipeline
Four specialized models work together: text extraction, vision analysis, OCR, and named entity recognition feed into a coordinator LLM that produces semantic classifications. All running locally.
Key Benefits
True zero-trust privacy
No file content, metadata, or thumbnails ever leave your machine for AI processing. Classification is completely local — no cloud, no exceptions.
Settings stay local
All settings, history, and configurations are stored locally in a SQLite database on your machine. We never store your files or documents.
No telemetry
We don't collect usage data, analytics, or crash reports. Your workflow is your business.
No accounts required for AI
No API keys, no cloud subscriptions, no sign-ups. Download Talyx, let it fetch models once, and classify files forever — even without internet.
Ready to organize your files?
Join the waitlist to be the first to try Talyx.