Concepts

Data Model & Identity

How Introspection organizes AI activity and ties telemetry to the right user, conversation, and agent.

Hierarchy


Organization
└── Project
    └── System / Service
        └── Agent
            └── Conversation
                └── Response
                    └── Feedback and events

Level	What it represents	Example
Organization	Team or company	Acme Corp
Project	Product, environment, or application	`support-assistant`, `staging-agents`
System / Service	Runtime emitting telemetry (from tracing service name)	`api-server`, `agent-worker`
Agent	Named AI unit handling work	An OpenAI Agent, a Claude session, a workflow component
Conversation	Multi-turn interaction from the user’s perspective	A chat thread, a task session
Response	One model output inside a conversation	A single assistant message
Feedback / Events	Signals attached to the user experience	Thumbs up/down, comments, product events

Conversation Resolution

Introspection resolves conversations in this order:

Explicit conversation ID — if your app sets gen_ai.conversation.id via the SDK, that value is used directly.
Previous response ID — if previous_response_id is set, the response is linked to the conversation that contains that prior response.
History-based matching — if neither is set, Introspection looks at recent activity from the same user and service to find a likely match.
New conversation — if no match is found, a new conversation is created.

If your app already has stable thread or session IDs, set them explicitly for the most reliable resolution.

Identity

Identity is what makes telemetry useful at the product level. It answers:

Which user was involved
Which conversation it belonged to
Which agent produced the response
Which feedback belongs to which answer

Types

Identity type	When to use	SDK method
User	Logged-in or stable users	`set_user_id()` / `withUserId()`
Anonymous	Pre-login, anonymous feedback, trial traffic	`set_anonymous_id()` / `withAnonymousId()`
Conversation	Multi-turn threads with a stable ID	`set_conversation()` / `withConversation()`
Agent	Multiple assistants or workflow components in one app	`set_agent()` / `withAgent()`

See the JavaScript, Python, or Rust SDK guide for code examples.

Baggage

The SDKs store context in OpenTelemetry baggage so spans, events, and feedback read it automatically within the same execution scope.

Baggage key	Set by	Meaning
`identity.user_id`	`set_user_id()` / `withUserId()`	Authenticated user
`identity.anonymous_id`	`set_anonymous_id()` / `withAnonymousId()`	Anonymous user
`gen_ai.conversation.id`	`set_conversation()` / `withConversation()`	Conversation
`gen_ai.request.previous_response_id`	`set_conversation()` / `withConversation()`	Link to a prior response
`gen_ai.agent.name`	`set_agent()` / `withAgent()`	Agent name
`gen_ai.agent.id`	`set_agent()` / `withAgent()`	Agent identifier