Version 2.19.0
Chats in projects, new models, usage limits, and more reliable documents
What's new
- Organise chats after the fact — You can now move existing chats into projects or remove them again. Older conversations stay neatly grouped by topic.
- Adjust previews more flexibly — The preview for code, images, and documents can now be shown wider or more compactly. Chat and content are easier to view side by side.
- New models for chat and image generation — DeepSeek R1, DeepSeek V3.2 and Kimi K2.5 are available as new text models. For images, the Flux 2 family replaces the older Flux Schnell and offers better quality and more variants.
- Freshservice as a new tool — Assistants can now work with Freshservice — search, create or update tickets and include relevant attachments.
- Clearer tool steps in chat — When an assistant uses tools, the chat shows clearer status messages such as search, file lookup or code execution. It's easier to follow what the answer is working on.
- Authors visible in sources — Documents from OneDrive and Google Drive now keep author information and show it in source references.
For administrators
- Usage limits per user group — You can manage monthly allowances for text, images, API usage, and related features. Users see their usage, get warnings, and can request an upgrade when needed.
- Manage sponsored access — Partner and sponsorship access can be managed in the admin area and automated through an external API.
- Guide model transitions — Models being phased out are clearly marked in the model picker. Users are guided toward current alternatives earlier.
- More transparent costs — Model prices and credit displays are shown more consistently in Euro; Gemini model accounting follows the current pricing structure.
- Allow email domains more flexibly — Subdomains are accepted more reliably when they belong to an allowed parent domain.
Fixes
- More stable document processing — Large or faulty files are less likely to stall processing. Uploads, synchronisation and search are more reliable, especially for larger document collections.
- DataVault counters are accurate again — Document and search counters now show real values instead of placeholders or conflicting numbers.
- More reliable assistant tools — External tools and database connections fail less often with unclear errors; Freshservice and Microsoft SQL queries are more robust as well.
- Fewer empty responses — Responses from certain Anthropic models are handled more reliably, so empty chat messages occur less often.