​

Proactive AI

​

What is Proactive AI?

• Monitors your integrations for relevant events
• Anticipates your needs based on context and patterns
• Acts autonomously when appropriate
• Learns from every interaction to improve over time

​

Core Capabilities

​

1. Event-Driven Intelligence

# Example: Email trigger detection (apps/api/app/agents/prompts/trigger_prompts.py)
# When an important email arrives, GAIA can:
# - Analyze the sender and context
# - Extract action items
# - Create calendar events
# - Draft responses
# - Update your todo list

• Gmail (new emails, labeled emails, matching filters)
• Google Calendar (event starts, event created, event updated)
• GitHub (PRs, issues, commits)
• Slack (messages, mentions, channels)
• LinkedIn (messages, connection requests)
• And 100+ more integrations via Composio

​

2. Context-Aware Actions

# From: apps/api/app/agents/prompts/workflow_prompts.py:358-440

# GAIA receives complete email context:
- From: sender@example.com
- Subject: "Urgent: Client Meeting Tomorrow"
- Content: Full email body with attachments
- Received: Timestamp

# And intelligently decides:
# - Should I create a calendar event?
# - Do I need to search for background info?
# - Should I draft a response?
# - Any todo items to create?

​

3. Autonomous Decision Making

​

Intelligent Workflow Execution

​

Trigger-Based Automation

# From: apps/api/app/models/workflow_models.py:18-30

class TriggerType(str, Enum):
    MANUAL = "manual"      # User-initiated
    SCHEDULE = "schedule"  # Cron-based timing
    INTEGRATION = "integration"  # Event-driven from connected apps

​

Learning System

​

Skill Learning

# From: apps/api/app/agents/memory/skill_learning/service.py:73-90

# GAIA learns procedural knowledge:
class Skill:
    trigger: str           # "When user asks to schedule a meeting"
    procedure: str         # "First check calendar, then search emails for context"
    tools_used: List[str]  # ["GOOGLE_CALENDAR_LIST_EVENTS", "GMAIL_SEARCH_MESSAGES"]
    what_worked: str       # "Checking calendar first prevents double-booking"
    gotchas: str           # "Always confirm timezone with user"
    optimal_approach: str  # "Use 'find_time' to automatically suggest slots"

1. LLM Extraction - Analyzing successful conversations
2. Self-Reflection - The executing LLM documents its own experience

​

User Memory

# Stored in mem0 with graph structure
- "User prefers morning meetings"
- "ClientX's CEO is Sarah Johnson ([email protected])"
- "Project deadlines are always Friday EOD"
- "User doesn't work Wednesdays"

​

Proactive Notifications

• In-app notifications (real-time updates)
• Email summaries (daily digests)
• Desktop notifications (important events)
• Mobile push notifications (urgent items)

• Priority Detection: Not all events deserve interruption
• Batch Summarization: Group related notifications
• Timing Optimization: Deliver during focus hours or breaks
• Action Shortcuts: One-tap to approve/dismiss/snooze

​

Privacy & Control

• All data processing happens in your account
• No data shared between users
• You can pause/resume proactivity anytime
• Full audit logs of all autonomous actions
• Granular permissions per integration

// Settings for proactive behavior
interface ProactivitySettings {
  enabled: boolean;                  // Master switch
  autonomyLevel: "low" | "medium" | "high";
  quietHours: { start: string; end: string; };
  allowedIntegrations: string[];     // Which apps to monitor
  autoApproveCategories: string[];   // Categories that don't need approval
}

​

Technical Implementation

​

Agent Architecture

# From: apps/api/app/agents/core/agent.py:38-114

async def _core_agent_logic(
    request: MessageRequestWithHistory,
    conversation_id: str,
    user: dict,
    user_time: datetime,
    trigger_context: Optional[dict] = None,  # Key for proactive execution
):
    # Build context with trigger data
    # Execute graph with full awareness
    # Store learnings for future improvement

​

Execution Modes

• Real-time user interactions
• Immediate feedback
• SSE-based communication

• Background workflow execution
• Batch processing
• API integrations

​

Best Practices

1. Define Clear Boundaries: Set up workflows for predictable scenarios first
2. Monitor & Adjust: Review autonomous actions regularly
3. Provide Feedback: Correct GAIA when it misinterprets context
4. Use Constraints: Set policies for when NOT to act autonomously
5. Test Incrementally: Enable one proactive feature at a time

​

Related Concepts

• Workflows - Automated action sequences
• Memory System - How GAIA remembers and learns
• Integrations - Connecting your digital ecosystem
• Agents - Specialized AI workers for different tasks

• Create your first automated workflow →
• Explore available integrations →
• Understand the memory system →