Skip to main content
The Sponsored Intelligence Protocol defines four tasks for managing conversational brand experiences:

Session Lifecycle

Two Entry Points

With pre-session lookup (dotted line): Host calls si_get_offering first to show products before consent. The offering_token carries what was shown into the session. Direct session (solid line): Host calls si_initiate_session directly. Brand agent shows products as part of the conversation and tracks them internally. Both are valid. Use pre-session lookup for sponsored search results where you want to preview products before asking the user to engage.

Tasks

TaskDescriptionInitiator
si_get_offeringGet offering details, availability, and matching products (anonymous)Host
si_initiate_sessionStart a conversation with a brand agentHost
si_send_messageExchange messages within an active sessionHost
si_terminate_sessionEnd the session with appropriate handoffEither

Transport Options

SI tasks work over both MCP and A2A protocols:

MCP Transport

{
  "method": "tools/call",
  "params": {
    "name": "si_initiate_session",
    "arguments": {
      "context": "User wants to fly to Boston next Tuesday morning",
      "identity": { /* ... */ }
    }
  }
}

A2A Transport

{
  "task": "si_initiate_session",
  "payload": {
    "context": "User wants to fly to Boston next Tuesday morning",
    "identity": { /* ... */ }
  }
}

Common Patterns

Minimal Session (No Identity)

For anonymous browsing without personalization:
{
  "context": "User interested in product information",
  "identity": {
    "consent_granted": false,
    "anonymous_session_id": "anon_xyz789"
  }
}

Full Identity Session

For personalized experiences with consented PII:
{
  "context": "User wants to book a flight",
  "identity": {
    "consent_granted": true,
    "consent_timestamp": "2026-01-18T10:30:00Z",
    "consent_scope": ["name", "email", "shipping_address"],
    "user": {
      "email": "jane@example.com",
      "name": "Jane Smith"
    }
  }
}

Campaign-Triggered Session

When SI is invoked as part of a media buy:
{
  "context": "User searching for flights to Boston",
  "media_buy_id": "media_buy_q1_promo",
  "placement": "chatgpt_search",
  "offering_id": "premium_upgrade_offer",
  "identity": {
    "consent_granted": true,
    "user": { "email": "jane@example.com" }
  }
}