What's new
What each update brings — only the changes you'll actually notice.
- v0.63.7Fixed
Models now start on computers that use a proxy, and fixed-name connections are more reliable.
- v0.63.6Improved
Local network access can use a fixed name, plus fixes for known issues.
Connect by a fixed name
Turn it on in Settings and other devices connect by a fixed name, so a changed network address never means a changed URL. The name prefix is yours to choose.
- v0.63.5Fixed
Fixes API endpoint issues and a settings page display issue.
- v0.63.3Improved
Hardware diagnostics show your graphics acceleration status and remind you when the driver needs updating.
- v0.63.2Fixed
Local network access is easier to use, and conversation length now fits your computer automatically.
Local network access
Turn it on and the addresses you can connect to are listed for you. These connections need no password, so use it only on networks you trust.
- v0.63.1New
Chat roles are yours to create and change.
Change a role any time
Change its name, description and wording without starting over. You can also copy a built-in role and adjust it.
See where your wording goes
The management view shows where your wording sits in the conversation.
- v0.63.0New
Conversations show how much room is left, and the settings explain themselves.
See the room you have left
A ring next to the input box shows how much this conversation has used. Start a new one when it gets close to full.
Model details are accurate
The length shown in the model list is the model's own limit, so you can pick by it directly.
Advanced settings warn you first
Every field states its default and range, and tells you the risk before you change it.
- v0.62.0New
Qwen 3.8 27B joins the model library.
Qwen 3.8 27B is available
Download it from the model page and it is ready to use. It reads images and handles very long documents, and like the others it runs entirely on your computer.
- v0.61.1New
Speech transcription supports more audio and video formats.
- v0.61.0Fixed
Improves speech transcription output.
- v0.60.2Fixed
The model list keeps up with new models on its own, and downloads say what they will fetch.
- v0.60.1New
Muse Glimmer 30B joins the model library.
Muse Glimmer 30B is available
Download it from the model page and it is ready to use. It reads images and reasons, and like the others it runs entirely on your computer.
Engine updates in one click
When an update is needed the app tells you first, and one click updates it and starts the model.
- v0.59.0Improved
Transcripts follow your interface language, and installation is steadier.
- v0.58.0Improved
Chat answers are more dependable, and image generation is steadier.
More dependable answers
Answers use the interface language you picked, and say so plainly when something is uncertain instead of inventing figures.
- v0.57.0New
Attach a document to a chat and ask about it.
Ask about a document
Attach a PDF, Word or text file and ask away, long ones included. Answers come with the passages they rest on.
Works for insights too
Results from text recognition and the speech workshop can be questioned the same way.
- v0.56.0New
Adds image upscaling, and image generation is faster.
Image upscaling
Upload a picture to enlarge it and bring detail back. No regenerating, no prompt needed.
- v0.55.0Improved
The chat and insight screens have been reorganised and read more easily.
Only updates that affect you are listed, so version numbers may skip.