v0.28.0

Added

• Thinking mode "Unlimited": no token limit for thinking, only a time limit. The limit can be set under Settings → LLM
  or with LLM_THINK_LIMIT_MIN (default 15 minutes, 0 = none) and applies to all thinking modes. It never cuts off the
  answer itself.

Changed

• If thinking reaches the time limit or uses up the tokens, the model now answers once without thinking instead of the
  run failing. The curator notes this as a warning in the run.
• The curator now tells the model the required date format and today's date for appointments.

Fixed

• Appointments created by the curator were missing from the calendar when the date was not in the exact format. Dates
  like "Tue, 03.11.2026" or "3 November" written in German are now converted, a missing start is taken from the
  appointment's text, and dated entries in a "## Termine" section of a source become appointments even if the model
  skips them.
• Appointments on different days could be merged into one because of similar or equal names, which would have changed
  the date. They now stay separate notes.