Document extraction picks up where it left off, with stronger connections
Document extraction now shows live progress and recovers results after connection interruptions. New Claude connections offer the required workflow permissions correctly, tunneled data sources connect without extra setup, and spoken presentations read numbers more naturally.
When extracting fields from an image or PDF document, the portal now keeps the job running in the background and shows live progress. Even if the connection drops briefly, Flow reads the saved status again and recovers the result. The previous PDF extraction method that could stall has been removed; if the image created from a document page is empty or excessively large, Flow stops the job instead of producing a meaningless result.
When a new Claude connection is created, the permission screen now offers the complete set of permissions to apply a draft and publish or run a workflow. An existing connection may need its connection record refreshed before these permissions appear. Designs prepared in a conversation can therefore go beyond viewing, within the permissions you grant. Tunneled data sources on private networks also work without extra connection setup; private addresses can be saved safely in the designer, and dashboards can access those sources.
In spoken dashboard presentations, money, percentages, dates, and large numbers are now read naturally in the response language, while the number formatting on screen stays unchanged.
In this release
- ImprovedLive progress and result recovery after reconnecting during document extraction
- FixedSafe stopping of empty or excessively large PDF page images during extraction
- FixedComplete draft-apply, workflow-publish, and workflow-run permissions in new Claude connections
- FixedTunneled data sources that work without extra connection setup
- ImprovedSpoken dashboards that read money, percentages, dates, and large numbers naturally in the presentation language