On‑Device AI:Private Workspace

What's new in 1.53.0 Adding your own GGUF model from Hugging Face is simpler. Chat Setup checks the model's chat template before you download it. When the template is compatible, Chat Format uses it automatically. If it isn't, you confirm Format Tokens first, so the model is ready to chat once the download finishes. You can switch between the two later without losing your manual settings. • OpenCode Go is now a built-in cloud provider. Add your Go key and choose a model. Your existing OpenCode Zen connection and saved selections stay as they are. • Connect to an Unsloth server from AI Chat. Enter its address, add an API key only if your server requires one, then pick a discovered model or enter its ID. • Conversation Mode shows what it is doing in one place: listening, responding, preparing speech, or speaking. Longer speech preparation shows elapsed time, and the microphone animation appears only while it is actually listening. • Context usage for local GGUF models is measured the same way the prompt is sent. For many popular model formats, the meter reads lower and more room is left for documents and knowledge. We've also made Continue finish cleanly when a model has nothing more to add, instead of starting the answer over, and kept Calculator available when you continue or regenerate an answer. Failed tool results now appear as warnings.