Fabric
Talk to SalesDownload
v1.64

v1.64: Push-to-Talk Dictation

|2 min read

What's New

  • Push-to-Talk Dictation: Dictate into any macOS app by holding a hotkey (default Fn), with audio captured from the moment you press so nothing is lost.

  • Remote Sessions: Connect to a live desktop chat from the Fabric CLI or a paired phone over your network, and approve permission prompts remotely.

  • Attachments Overhaul: File size limits and project copies are removed, large files process in the background, PDFs read by page range, and duplicate images are stored once.

  • Deep Research Command: A new guided research workflow fetches web pages and converts them to clean text for the agent to analyze.

  • Built-In Agent Skills: Document skills for Word, PowerPoint, Excel, and PDF now ship with the app, plus a deslop skill that removes AI-writing patterns from prose while preserving facts and your voice.

  • Workflow Resume Banner: Reopening a chat with an unfinished workflow shows a banner offering the next step, dismissible per chat.

  • Divider Drag-to-Close: Dragging the file navigator divider past its minimum width closes the panel; dragging back reopens it in the same gesture.

Bug Fixes

  • Complex tasks now survive temporary connection failures and retry automatically instead of stopping.

  • Conversation summaries no longer get truncated when the connection drops — failed summaries leave the chat unchanged and can be retried.

  • Fixed a crash loop caused by malformed AI responses.

  • Fixed the Write tool writing placeholder text instead of actual content.

  • Previously activated tools no longer disappear when reopening a chat.

  • Scheduled routines that were missed while Fabric was closed now catch up reliably at launch instead of failing or running in the wrong project.

  • Routines are now fully localized in all six supported languages.

  • Updating on macOS no longer gets stuck in a loop, and the download progress window is no longer hidden behind other windows.

  • AI text now appears much faster and no longer keeps typing after the response finishes.

  • Voice setup is improved with clearer permissions, live transcripts as you speak, and fixes for retrying and restarting onboarding.