Docs/Intelligence

Ambient Intelligence

Overview

The ambient intelligence system observes your activity across channels, connectors, and behaviors to build a model of your routines. It detects patterns, anticipates needs, generates personalized briefings, and delivers notifications at the right priority level. Unlike behaviors (which you configure explicitly), ambient intelligence emerges from observation -- it learns what matters to you and surfaces it before you ask.

How It Works

Auxiora's ambient engine runs a continuous observation loop:

Observe -- Every interaction (messages, connector events, behavior executions, calendar changes) is recorded as an ObservedEvent with a timestamp and metadata.
Detect -- A sliding-window pattern engine (default: 7-day window) analyzes events for recurring schedules, frequency patterns, and temporal correlations. Patterns require at least 3 occurrences and 30% confidence to be surfaced.
Anticipate -- Detected patterns feed into the anticipation engine, which predicts upcoming needs and suggests proactive actions.
Notify -- Results are routed through a priority-based notification system that batches low-priority items and escalates urgent ones.

Features

Pattern Detection

The pattern engine detects four types of patterns:

Pattern Type	What It Detects	Example
Schedule	Events recurring at similar times of day	"Code review events frequently occur around 14:00"
Preference	Events occurring at regular intervals	"Email checking occurs roughly every 2.5 hours"
Correlation	Event B frequently follows event A within 5 minutes	"Slack messages often follow GitHub notifications"
Trigger	Event sequences that signal a specific condition	"Three failed CI runs in a row"

Each pattern includes a confidence score (0--1), evidence trail, occurrence count, and timestamps for when it was first detected and last confirmed. Pattern confidence increases by 0.05 each time the pattern is re-confirmed.

Briefings

Configurable morning and evening summaries that pull from all connected data sources.

Morning briefing sections:

Pending notifications
Today's calendar schedule
Active tasks
Observed patterns (50%+ confidence)
Anticipated upcoming events

Evening briefing sections:

Pending notifications
Tomorrow's schedule
Active tasks
Patterns and anticipations

Briefings are configurable:

{
  "ambient": {
    "briefings": {
      "enabled": true,
      "morningTime": "08:00",
      "eveningTime": "20:00",
      "categories": ["calendar", "tasks", "weather", "news", "patterns"],
      "maxItemsPerSection": 5
    }
  }
}

Generate a briefing on demand via CLI:

auxiora ambient briefing               # Morning briefing (default)
auxiora ambient briefing --time evening # Evening summary

Anticipation

The anticipation engine predicts what you will need next based on detected patterns:

Schedule-based -- If you always review PRs at 2pm, Auxiora prepares a PR summary 15 minutes before.
Frequency-based -- If you check email every 2.5 hours, Auxiora pre-fetches an email summary when the interval approaches.
Correlation-based -- If Slack activity always follows GitHub notifications, Auxiora prepares Slack context when a GitHub event arrives.
Trigger-based -- If a trigger pattern fires (e.g., repeated CI failures), Auxiora immediately generates a diagnostic summary.

Anticipations require a minimum confidence of 0.4 from the source pattern. Confidence is further scaled per pattern type (schedule: 0.8x, preference: 0.7x, correlation: 0.6x, trigger: 0.9x) to reflect prediction reliability.

Quiet Notifications

Not everything deserves an interruption. The notification system uses three priority levels:

Priority	Behavior	Use Case
alert	Immediate delivery, interrupts current activity	Production outage, urgent email from boss
nudge	Delivered at next natural break point	PR approved, meeting in 15 minutes
whisper	Batched into daily/weekly digest	Pattern detected, low-priority update

Notifications are sorted by priority within the queue. Old dismissed notifications are automatically pruned after 24 hours.

Configuration

CLI Commands

auxiora ambient status          # Show observed events, detected patterns, pending notifications
auxiora ambient briefing        # Generate a personalized briefing
auxiora ambient patterns        # Show detected behavioral patterns
auxiora ambient patterns -l 20  # Show up to 20 patterns
auxiora ambient notifications   # Show pending quiet notifications

Enabling / Disabling

Ambient intelligence can be toggled globally:

auxiora ambient enable          # Enable ambient features
auxiora ambient disable         # Disable ambient features

When disabled, no events are observed and no patterns are detected. Existing patterns and notifications are preserved and resume when re-enabled.

Dashboard

The Settings > Ambient page provides toggles for:

Pattern detection (on/off)
Briefing schedule (morning/evening times)
Briefing categories (calendar, tasks, weather, news, patterns)
Notification priority thresholds
Maximum items per briefing section

Architecture

The ambient system is composed of several cooperating modules:

Module	Responsibility
`AmbientPatternEngine`	Sliding-window event observation and pattern detection
`AnticipationEngine`	Predicts user needs from detected patterns
`BriefingGenerator`	Compiles data sources into structured morning/evening briefings
`QuietNotificationManager`	Priority-based notification queue with dismissal and pruning
`AmbientScheduler`	Coordinates timed execution of ambient tasks
`NotificationOrchestrator`	Routes notifications to appropriate delivery channels
`AmbientAwarenessCollector`	Gathers events from channels, connectors, and behaviors

All state can be serialized and restored across restarts via AmbientPatternEngine.serialize() and AmbientPatternEngine.deserialize().

Use Cases

1. Executive Assistant

The ambient system observes your calendar, email, and task management tools throughout the week. Each morning at 8:00 AM, it delivers a briefing with today's meetings, flagged emails, overdue tasks, and any patterns it has noticed (e.g., "You tend to schedule deep work on Wednesday afternoons"). Before each meeting, it prepares a context brief with attendee backgrounds and relevant documents.

2. Developer Awareness

As you work, the ambient system tracks GitHub notifications, CI pipeline results, and Slack activity. It detects that your CI pipeline has failed 3 times today and proactively surfaces a summary of the failures as a nudge notification. It also notices that you always check Slack within 5 minutes of a GitHub notification and begins pre-loading relevant Slack threads.

3. Meeting Intelligence

The anticipation engine detects a recurring pattern: you have a standup at 9:30 AM every weekday. Fifteen minutes before, it prepares context -- yesterday's standup notes, your completed tasks since then, and any blockers from your project tracker. The preparation is delivered as a nudge notification so it is ready when you need it, without interrupting deep work.