November 5, 2026 · 3.0.0-beta.18
Your own models: trained on your day, wherever the model lives.
Train on any computer you own, connect the open-weight models you already run, let your models learn from your day if you choose, and ask quick chat about your screen.
- New
Train on a computer you own: pair a Mac or Linux box that has the model, then train, score and serve on it from your laptop. Traffic is encrypted with a key derived from its pairing code.
- New
Host training on this computer for your others, or run velora train-host on a Linux box with MLX.
- New
Learn from your day, off until you turn it on: the app in front and its window title become a daily journal in Sources, and documents in folders you choose are imported when they change. Pause it, or delete everything it captured.
- New
Keep personal details for a model you own, or take them out first, when you create a dataset or prepare a run. Detected credentials always come out.
- Improved
Models on a computer you paired see personal details but not detected credentials, and Local only projects stay off them.
- Improved
Qwen, Llama, Gemma, Mistral and SmolLM models show their family's logo across Intelligence.
- New
Open-weight models you run are found and connected in one click: Ollama, LM Studio, llama.cpp, vLLM, Jan and MLX on this computer, with no API key. A server on another computer on your network connects without a key too.
- Improved
Open-weight models and API connections answer chats. They were connected but missing from the chat model menu; models you run yourself now come first.
- New
Quick chat (⌥Space) lets you choose the model and how hard it thinks, take a screenshot (or drag a region on a Mac) and record your screen. Models see the picture, or frames from the recording.
- New
Private mode asks your operating system to leave Velora's windows out of screen sharing and recordings. On Windows that covers every share; on macOS 15 and later a full-screen share can still show it, so share a window instead.
- Fixed
Pressing ⌥Space a second time puts the quick chat away and returns you to the app you were using, instead of bringing up Velora's main window.