AI Personal Assistant

The AI Personal Assistant (PA) gives every employee their own dedicated AI agent that monitors their inbox, tracks goals, drafts replies, conducts research, and manages projects, all personalized to their working style. Unlike standard AI agents that operate at the organizational level, the PA is scoped to an individual employee and adapts to their specific needs, preferences, and communication style.

What the Personal Assistant Does

The PA functions as an AI-powered personal productivity assistant that:

• Manages Email - Monitors the employee's shared mailbox and (optionally) their primary inbox, drafts replies to routine messages, categorizes and prioritizes incoming mail
• Tracks Goals - Takes employee-defined goals and autonomously breaks them into projects and tasks
• Conducts Research - Researches topics on demand and delivers concise summaries with key findings
• Manages Projects - Creates projects, tracks tasks, and reports progress toward goals
• Prepares Meeting Briefings - Reviews the employee's calendar and prepares context and notes for upcoming meetings
• Communicates Proactively - Wakes up on a configurable schedule (1-4 times per day) to work on assigned goals and check in with the employee
• Responds in Real Time - Reacts immediately to incoming emails at the PA's shared mailbox, chat messages, and manual wake-up requests
• Communicates with Management - Sends messages to the employee's manager on their behalf when escalation, status updates, or coordination is needed

How the PA Works

The PA operates on a dual execution model:

1. Scheduled Wake-Ups (1-4 times per day) - The PA proactively works on goals during configured working hours. Each wake-up follows a structured cycle: load context, assess current state, plan the session, execute work using assigned tools, communicate results, and log activity. Scheduled wake-ups are evenly distributed across the employee's working hours.

2. Real-Time Triggers - The PA responds immediately to incoming emails at its shared mailbox, chat messages from the employee, manual "Wake Up Now" requests, and goal status changes. These do not count against the scheduled wake-up frequency but do count toward the daily execution limit. Email triggers process new messages independently outside of scheduled cycles.

Execution Cycle

During each wake-up, the PA:

1. Loads its configuration, employee profile, active goals, and recent context
2. Assesses the current state (new emails, calendar events, pending tasks, overdue items)
3. Plans and prioritizes work based on the trigger type and active goals
4. Executes work using its independent tool pipeline (drafting emails, conducting research, updating projects)
5. Communicates results to the employee via the configured channel
6. Logs all activity and generates a summary if configured
7. Returns to sleep until the next scheduled wake-up or real-time trigger

Independent Tool Pipeline

The PA operates with its own independent tool pipeline, separate from other agent types. This pipeline includes:

• 16 inherited tools from AICOS: project management, task tracking, email, knowledge, reminders, goal management, and more
• 1 PA-specific tool: send_message_to_boss for escalating to the employee's manager
• PA-scoped data isolation: Six inherited tools have PA-specific overrides ensuring each PA only accesses its own employee's goals, projects, and email, not organization-wide data

The PA also injects admin-approved tenant tools dynamically at execution time based on the employee's tool assignments.

info

The PA extends the AICOS (AI Chief of Staff) executor architecture with individual-scoped goals, projects, tasks, and activity logging. Its separate execution queue and processor allow independent scaling and monitoring independent of AICOS execution.

Key Benefits

Individual Productivity Boost

Each employee gets a dedicated AI assistant that handles email triage, drafting replies, research, meeting preparation, and follow-up tracking, freeing hours each day for higher-value work. The PA adapts to the employee's communication style and preferences over time.

Two-Tier Safety Model

Administrators control which tools and AI models are available to each employee's PA. Employees then choose from that approved pool during their own setup. No one can escalate permissions through the PA, ensuring organizational security while giving employees flexibility.

Intelligent Onboarding

When a new PA is activated, the employee's velocity choice determines the onboarding approach:

• Ease Into It - Gradual introduction with weekly check-ins to understand work patterns, goals, and preferences
• Hit the Ground Running - Immediate full activation with the employee's configured settings; minimal hand-holding
• Live Chat - Synchronous onboarding session via the PA Hub chat interface for quick setup and immediate answers

All velocity paths culminate in a fully configured PA ready to execute work on the defined schedule.

Full Personalization

Employees name their assistant, choose an avatar, set a communication style (Professional, Casual, Concise, or Detailed), and provide custom instructions that shape every interaction. The PA adapts its tone, email signatures, and response style to match these preferences.

Configurable Privacy

Three privacy levels (Private, Metadata Only, Full Visibility) let organizations balance employee autonomy with administrative oversight. The privacy policy is set at the tenant level and applies uniformly to all PAs. It controls only what administrators can see, not what the PA can do. This makes the PA suitable for high-trust environments and regulated industries alike. See Privacy & Governance for details.

Proactive Goal Tracking

Unlike reactive chatbots, the PA wakes up on a schedule to proactively work on assigned goals, track projects, and report progress. Employees define what they want help with, and the PA takes independent action within the boundaries set by the employee and administrator.

Manager Communication

The PA includes a dedicated send_message_to_boss tool that lets it escalate issues, provide status updates, or coordinate with the employee's direct manager on their behalf. Messages are sent via email from the PA's shared mailbox to the manager's email address (resolved from Microsoft Entra) and stored in a conversation record for history tracking. The tool respects EscalationBehavior settings: at tier 3 (live_chat), the PA asks the employee before escalating; at tiers 1-2, it can escalate autonomously based on the situation.

Real-Time Updates

The PA Hub receives real-time updates via SignalR when the PA starts or completes work, finishes tasks, changes status, or sends chat messages. Employees see live indicators on the dashboard without needing to refresh the page. The usePASignalR hook manages user-scoped event groups so each employee only receives updates for their own PA.

Configurable Summaries and Notifications

Employees choose how they want to stay informed: daily summaries, per-cycle summaries, or no summaries at all, delivered via email, in-app notifications, or both.

How the PA Differs from Standard Agents and AICOS

Aspect	Standard AI Agents	AICOS (AI Chief of Staff)	Personal Assistant
Scope	Organization-wide email processing	Strategic goals for the business	Individual employee productivity
Trigger	Incoming emails	Scheduled daily wake-up	Scheduled wake-ups + email + chat triggers
Controlled By	Administrator	Business Owner	Employee (within admin boundaries)
Goals	N/A (reactive)	Strategic business objectives	Personal productivity goals
Memory	Contact-specific preferences	Organization-wide knowledge	Employee-specific context and profile
Personalization	Agent instructions only	Communication preferences	Full (name, avatar, style, custom instructions)
Privacy	Admin has full visibility	Admin has full visibility	Configurable privacy policy (3 levels)
Data Access	Shared mailbox only	Shared mailbox only	Shared mailbox + optional primary mailbox, calendar, contacts, files, meetings
Tool Pipeline	Standard tenant tools	CAIOO tool set	Independent pipeline (17 tools with PA-scoped data isolation)

PA-Specific Data Access

The PA can access the employee's Microsoft 365 data based on permissions the employee grants during setup. These permissions are tenant-specific and controlled entirely by the employee:

Permission	Options	What It Enables
Primary Mailbox	No access / Read-only / Read & write	PA monitors the employee's personal inbox, categorizes messages, drafts replies (saved as drafts by default). Only affects actions within the employee's personal inbox; email triggers on the PA's shared mailbox are independent.
Calendar	On / Off	PA reads calendar events for scheduling context, meeting preparation, and conflict checking
Contacts	On / Off	PA accesses the employee's contact list for context and resolution
OneDrive	On / Off	PA reads files from the employee's OneDrive for Business
Meetings	On / Off	PA reads Teams meeting transcripts to extract action items and tasks (off by default)

These permissions are controlled entirely by the employee and can be changed at any time from PA Hub > Settings. The privacy policy at the tenant level does not restrict what data the PA can access, only what administrators can see about PA activity.

Use Cases

Executive Assistant

A VP of Sales names their PA "Friday" and assigns goals for inbox management, daily briefings, and meeting preparation. Friday wakes up twice a day, reviews the VP's inbox, drafts replies to routine messages, prepares briefing notes for upcoming meetings, and sends a daily summary of completed work. When an urgent client issue arises, Friday uses send_message_to_boss to escalate directly to the VP's manager.

Research Support

A product manager assigns their PA the "Research Assistant" and "Knowledge Curator" goals. The PA autonomously researches competitor pricing, summarizes industry reports, and organizes findings into a personal knowledge base that the PM can reference anytime.

Administrative Efficiency

An HR director uses their PA to track follow-ups on employee onboarding tasks, draft responses to routine policy questions using the company knowledge base, and maintain a running project for quarterly compliance reviews.

Writing and Communication

A marketing manager configures their PA with a "Casual" communication style and assigns a "Writing Assistant" goal. The PA helps draft blog posts, social media content, and internal communications, matching the manager's preferred tone and always including the standard company sign-off.

Getting Started

Setting up a Personal Assistant is a two-phase process:

1. Admin provisions the PA - Navigate to Manage > Team > Users, select an employee, assign the PA User role, and complete the Admin Setup Wizard
2. Employee configures the PA - The employee receives a setup email, opens the PA Hub, and completes the Employee Setup Wizard to personalize and activate their assistant

Once activated, the PA begins a structured onboarding period where it learns the employee's work patterns and preferences before transitioning to full operational mode.

PA Status Lifecycle

The PA moves through the following statuses during its lifecycle:

Status	Meaning
Pending Admin Setup	Admin has assigned the PA User role but has not completed the admin wizard
Pending Employee Setup	Admin wizard complete; waiting for employee to activate
Active	PA is running on schedule and processing triggers
Paused by Employee	Employee has paused the PA; employee can resume
Paused by Admin	Admin has paused the PA; only admin can resume
Deactivated	PA has been permanently stopped; requires full re-setup to reactivate

Components Overview

Component	Description
Admin Setup Wizard	Provision a PA for an employee (shared mailbox, tools, models, settings)
Employee Setup Wizard	Personalize and activate the PA (profile, goals, tools, name, schedule)
PA Hub Guide	Navigate the PA's dashboard, goals, projects, chat, activity, insights, and settings
Privacy & Governance	Privacy levels, admin controls, pause/resume, deactivation, and audit trail

Related Topics

• AICOS Overview - The organizational-level AI executive that the PA architecture builds on
• AI Configuration - LLM provider setup required before provisioning
• Agent Intelligence - Memory and artifact features available to all agents