Nexmi - AI Anticipatory Task and Memory Assistant

Nexmi is an AI-powered personal task and memory assistant built around a single idea: instead of asking the user when to be reminded, it works out when a reminder will actually help.

A traditional reminder app waits to be told a time. Nexmi takes a plain sentence, understands the deadline and everything that has to happen first, scores how much each task is at risk of being missed, and only then decides whether to speak up — so the reminders that do arrive are worth reading.

Write It The Way You Would Say It

There is no long form to fill in. The user types one sentence and the assistant reads it back as a structured task for confirmation before anything is saved.

• Natural-language capture from a single input box
• Deadline extraction from phrases like "Monday", "tomorrow evening" or "before Friday"
• Prerequisites detected automatically and linked as dependencies
• Importance classification, with a manual override that locks the value
• Recurring patterns recognised from wording such as "every evening"
• The original sentence is kept on the task as "in your words"
• Nothing is invented — if no time was given, none is assumed

Anticipation Engine

The core of the product is a deterministic scoring engine in PHP. The language model handles understanding; the business rules, dates, scoring and scheduling stay in application code.

• Every pending task receives an anticipation score out of 100
• Weighted inputs: deadline proximity, importance, unfinished dependencies, forgetfulness risk and past behaviour
• Configurable thresholds decide whether a task is watched, scheduled, reminded or treated as urgent
• A task existing is never enough on its own to trigger a reminder
• Scores and the next recommended action are visible on the task screen

Dependency Awareness

Dependencies are treated as first-class data, not notes. A deadline with an unfinished prerequisite is the clearest signal that something is about to go wrong.

• Prerequisite steps can be added, ticked off or removed at any time
• Task screens show how many dependencies are still pending
• The blocking step is surfaced as the next recommended action
• Clearing the blocker immediately drops the risk and the warning with it

Reminders That Explain Themselves

Reminders are queued by a background job rather than fired on a fixed timer, and each one carries the reason it exists.

• Due now, scheduled and history views of the reminder queue
• Trigger reasons such as dependency, deadline or anticipated risk
• A plain-language "Why this matters" explanation drawn from the user's own data
• Done, I forgot, Snooze and Dismiss actions on every reminder
• Full lifecycle tracking from scheduled through sent to acted upon

Behaviour Learning & Notification Fatigue

The assistant records how the user responds and adjusts how hard it pushes, while strictly limiting how often it is allowed to interrupt.

• Completions, snoozes, dismissals and missed tasks recorded per user
• Forgetfulness risk raised for the kinds of task that repeatedly slip
• Maximum reminders per day and a minimum gap between them
• Quiet hours during which nothing is delivered
• Urgent items can override the limits when genuinely warranted
• Dismiss is treated as a signal to stop, not as a completion

AI Assistant

A dedicated assistant screen answers questions about the user's own workload — what is forgotten, what to do first, what is due this week, what is blocked.

• Answers are grounded strictly in saved tasks; the model may not invent items
• A short written summary followed by items ranked by urgency
• Each line explains itself and links straight to the task
• Preset questions plus free-text asking

Recurring Tasks

Completing or missing an occurrence never ends the series. The next instance is created with its own deadline, and a missed occurrence is closed as missed rather than left to nag.

Reminders On The Phone

Nexmi is installable and delivers real notifications to a phone browser with no app store involved.

• Standard Web Push with VAPID keys and a service worker
• Per-device subscribe and unsubscribe with a test notification
• Progressive web app manifest and offline shell
• Notification preferences and quiet hours in settings

Background Jobs

Three idempotent HTTP cron endpoints, token protected, drive everything that happens without the user present: anticipation scoring, reminder processing and recurring task rollover. Each run is logged so repeated execution is always safe.

Accounts & Security

• Email signup confirmed by a time-limited six-digit code
• Session-based authentication with CSRF tokens on every write
• Rate limiting on authentication and AI endpoints
• Per-user data isolation on every query
• Prepared statements throughout and hashed passwords
• Timezone-aware handling, with UTC as the single source of truth

Platform Highlights

• Anticipatory reminders instead of fixed alarm times
• One-sentence natural-language task capture
• AI extraction of deadlines, prerequisites, importance and repeats
• Weighted anticipation score with configurable thresholds
• Dependency chains that change what gets escalated
• Every reminder explains why it was raised
• Behaviour learning and forgetfulness risk
• Strict notification fatigue controls and quiet hours
• "What am I forgetting?" AI assistant grounded in real data
• Recurring tasks that roll forward on completion or a miss
• Web Push notifications and installable PWA
• Token-protected idempotent cron endpoints
• Deterministic business logic kept out of the language model
• Responsive, restrained productivity-app interface