AutoScheduler MAX — LLM picks the anchors, you keep control
A second-tier scheduler that uses an LLM to plan your week with judgment. The regular auto-scheduler reflows around what MAX picks. 30 runs per month included.
- AddedMAX panel — click the wand button in the toolbar. Watch the model think live, then approve, refine, or discard. The greedy scheduler keeps doing its thing around the anchors MAX places.
- AddedStreaming reasoning. The model's thinking and text appear as they're produced — no 25-second blank wait.
- AddedProvider choice — OpenAI's GPT-5.5 (default) or Claude Sonnet 4.6 with extended thinking. Both render the same anchors.
- AddedFive plan modes — Default, Catch up, Deep focus, Launch, Recovery — each tunes the strategy for a different kind of week.
- AddedRefine without burning a run. Type "push deep work to mornings" and re-stream for free, up to 5 times per plan.
- AddedCalendar overlay preview. Proposed anchors render directly on the calendar with a dashed amber border. Click a card and the calendar jumps to that anchor.
- AddedUndo applied. A toast with 10 minutes to roll back a MAX apply if you change your mind.
- AddedHistory view in Settings → MAX. Last 10 runs with reasoning, what you kept vs discarded, and any refinements.
- AddedIterative learning. MAX reads your past 5 runs to surface patterns ("tends to keep morning workouts, discards Friday-PM deep work") and self-tunes future plans.
- AddedStyle note + privacy toggle in settings. One textarea describes how you like to work; one switch redacts task titles before they're sent to the LLM.
- ImprovedHard validation. Anchors that fall outside your working hours, overlap a meeting, land after a project deadline, or sit before a task's start moment get rejected with the reason shown.
- ImprovedPre-aggregated analytics in the prompt — peak hours, peak weekdays, average session length — so MAX picks slots that match how you actually work.