Changelog
Start typing to search…

Releases


By October 29, 2025

General

  • A dynamic progress bar above the chat input now shows how much of the AI’s context window is currently being used:
    • It updates in real time based on the conversation history and any files uploaded but not yet submitted, helping you manage context limits more easily.
    • Move your mouse over the bar to see a breakdown of the current use.
    • The progress bar turns yellow and then red when the context window approaches the model's limit.
    • Clicking on "New chat" will clear the context window and start a new chat.
    • Note that every model has a different-sized context window. Mistral 3.2 is the recommended current model for most tasks.
  • Global and Company Assistants can be duplicated to create a copy and allow you to run your personal version of it.
  • Admins can turn their Personal Assistants into Company Assistants to make them available to everyone.
  • Tables and Excel sheets are converted more efficiently into tokens and therefore take up less space in the context window.
  • Chat names can be edited via the three dots next to the name.

Helpful ressources

Models

  • When creating assistants, the best model (currently Mistral 3.2) is selected by default.
  • Recommendation: update your assistants to the model Mistral 3.2 for the bext performance.

Bugfix

  • When duplicating assistants, demo prompts are now being copied as well.
  • Fixed a bug that limited file upload to 1MB.
  • Miscellaneous bugfixes in the Gateway.

By October 8, 2025

General

  • Easier way to select Models and Assistants in the top menu
  • Maximum upload size for files specified in the upload tooltip

Helpful ressources

Models

  • Default when you log in for the first time changed to the strongest model, Mistral 3.2
  • Model explanatory texts updated, with recommendations for the strongest models

Knowledge

  • Assistants will now use their files by default, unless you deactivate the “Use Knowledge” button
  • A reminder now appears when deactivating “Use Knowledge”

Bugfix

  • “Ask in new chat" error fixed

By October 1, 2025

  • Improved rendering of code, links, and Markdown in chat
  • Setup of many user accounts via CSV import
  • Function to edit previous chat messages and restart the chat from there
  • Improved display of chat field and action buttons
  • Better error tracking
  • Links to privacy policy and legal notice pages

For API customers:

By September 4, 2025

🤖 AI model upgrades

  • Release of Mistral 3.2 with a 128k-token context window (four times larger than Gemma 3 or Llama 3)
  • Increase of maximum response size to 4,000 tokens (four times compared to before)
  • API customers: faster, better responses across all models when using structured-output specifications

🔑 More convenience & control

  • API keys can now be renamed at any time – for better clarity.
  • All chats can be deleted at once.
  • Requests can be continued via “Ask in new chat.”
  • Automatic following of responses is now configurable.
  • Dark/Light mode now automatically follows the system setting.
  • Code blocks can be copied with one click.

📂 Improved work with documents

  • Assistants can now be equipped with uploaded documents and thus have permanent access to extensive content.
  • More stable processing thanks to downstream embeddings and clearer error messages.
  • File size and type are displayed in the knowledge base.
  • Linked files can be downloaded directly from the chat.
  • Bugfix: issues embedding large files have been resolved.

⚙️ Stability & monitoring

  • Automated monitoring of LLMs ensures higher reliability.
  • More precise error messages when uploading images.
  • Bugfixes for assistant welcome messages and for usage tracking in streams.