
Nothing new to click this time. What this update removes are the dead ends: the long document that ran out of pages, the turn that stalled, the session that quietly reset. These are the ones you were most likely to run into.
Whole-document mapping keeps going past 480 pages, without rereading earlier windows to reach the rest. Coverage is preserved across calls, so the last page is as reachable as the first.
When a file is too dense to read in one pass, the read now tells you which page range to pick up from next, a long document can be walked to the end instead of truncating there.
Attached documents show how many pages the run expects to read and how it will pace them, replacing a context meter that never matched the work.
Creating or revising a very large document now returns a recoverable error you can act on, rather than a failed run.
A turn that stops making progress is cut short and retried, instead of hanging until it times out.
A model call that is truncated is bounded and recovered rather than reported as a failure.
When a run has data and tools available, the first turn is required to use them.
If a call fails outright, the run closes with the best answer from the work already done, or an explanation of what went wrong, instead of ending silently.
Shared-project members can download an AI-generated document with its contents intact, instead of an empty file.
The workspace you selected stays selected across browser sessions.
An expired session asks you to sign in again, rather than returning a server error or resetting you to your personal space.
Switching between document versions moves between versions that already exist, instead of creating new ones.
An uploaded image, a logo, a chart, can be reused in a document Tresor generates, and survives code-based edits that rewrite the rest of the file.
🔒 Privacy: None of this changes how your data is handled. Documents and attachments are still read only inside the sealed environm