What's new

A small follow-up focused on the two things you told us about most.

Improved

• The nudge agent now understands idle stretches and adjusts its tone instead of treating every gap as drift.
• Duplicate-detection on new tasks and goals is more forgiving of rephrasings.

Fixed

• Changing a task’s status by voice now reflects on the board immediately, no refresh required.
• Reduced background CPU use from the activity watcher on Windows.

The first public release of Work Assistant is here.

What’s in 1.0

• Voice chat — push-to-talk from the app, streamed live. Speech-to-text, an agent turn that can create and update your tasks and goals, and spoken responses back, sentence by sentence.
• Tasks & goals — a kanban-style board with comments, all manageable by voice or by hand.
• Proactive nudges — a scheduled midday check-in, plus activity-aware nudges that compare your recent window activity against what you said mattered, and speak up when they disagree.
• Consent-first monitoring — activity capture stays off until you accept a plain-language, versioned disclosure. Change what’s captured and everyone is asked again. Export or delete your data at any time.
• Bring your own model — every model the app uses is declared in one config file. Switch the whole stack between OpenAI and a self-hosted, OpenAI-compatible server with a single change.

Available for Windows and Linux.