Most AI tools send everything to someone else's server. Your drafts, your notes, your half-formed ideas — all routed through a cloud you do not control. Glassy does the opposite. Local AI is the default. Cloud AI is opt-in, explicit, and only for the work that needs it.
What runs locally
Qwen 2.5 (0.5B, 1.5B, and 3B variants) runs inside your browser via WebGPU. Grammar fixes, tag suggestions, summaries, and brainstorming — all on your machine. Whisper Base and Small handle voice transcription locally. Your audio never leaves the browser. These models download once and cache in IndexedDB. After that, they work offline.
When cloud AI gets involved
When a query needs a larger context window, harder reasoning, or a model you have not downloaded — Glassy routes to your configured cloud provider: OpenAI, Anthropic, or any OpenAI-compatible endpoint. This is opt-in. You choose when to send something. Cloud AI is reserved for the work the local model cannot handle. Every free account gets local AI for life. Pro accounts include cloud AI usage with top-up packs available.
The boundary is clear
We do not train models on your content. We do not sell your data. Local prompts and context stay local. Cloud requests only leave when you choose that path. The product is designed so privacy-sensitive work never needs to touch a server.