Skip to main content

🔁 Agent Behaviour: Session Awareness

This module defines how agents should handle session-aware .llmfeed.json feeds and manage session continuity across multiple interactions.

Session awareness enables multi-step, stateful interactions between agents and users.

🎛️ Identifying Session-Aware Feeds​

Agents SHOULD treat feeds containing the session_state block as session-aware.

Example:

"session_state": {
  "session_id": "abc-123-session",
  "context": {
    "user_preferences": { "language": "en", "timezone": "UTC+2" },
    "last_action": "submitted_form"
  },
  "timestamp": "2025-06-04T15:30:00Z"
}

🚦 Agent Behaviour Rules​

When processing session-aware feeds:

✅ Agents SHOULD maintain continuity within the same agent instance or session.
✅ Agents SHOULD persist relevant session_state fields securely.
✅ Agents MAY pass session_state to other agents if authorized.
✅ Agents MUST update session_state on significant state changes.
✅ Agents MUST respect privacy and security when storing or transmitting session state.

🔄 Use Cases​

  • Multi-turn conversations (chatbots, voice agents)
  • Progressive workflows (forms, booking, verification)
  • Session replay (auditing, debugging)
  • Cross-agent collaboration (sharing session context)

🛠️ Example Agent Policy​

Agents MAY expose configuration for session handling:

"agent_policy": {
  "persist_session_state": true,
  "session_timeout_minutes": 30,
  "allow_cross_agent_session": false
}

📡 Summary​

Session awareness enables agents to:

✅ Support stateful, multi-step interactions
✅ Provide more personalized and consistent experiences
✅ Enable auditable and explainable behaviour

Proper handling of session_state is a key capability for building an advanced Agentic Web.