Conversation Management

Manage agent interaction data through conversation APIs to build persistent memory in Data 360. Conversation APIs support the full lifecycle of an agent interaction, starting from creation through message ingestion to session closure. They also handle multi-agent sessions, agent handoffs, and long-running interactions.

Note: Conversation management is API-first. You perform all operations through the conversation APIs. For endpoint details, request schemas, and complete field reference, see Conversation Ingest APIs.

Conversation Lifecycle

Data 360 conversations progress through these stages.

Create a conversation: Establishes a new conversation entity. You provide a unique conversation ID, ownership information (who the conversation belongs to), the owner type (individual or employee), and optional metadata such as a display name and communication channel.
Create a session: Opens a session within the conversation. You provide a session ID, owner information, and at least one initial participant. A conversation can contain multiple sessions to support handoffs and pauses.
Add participants: Associates additional individuals, agents, or users with an active session. Each participant has an ID, type (individual, employee, or agent), and a role (owner or participant).
Ingest entries: Sends ordered sequences of interaction records to Data 360 as the conversation progresses. This is the primary mechanism for getting conversation data into the memory pipeline.
Close the session: Signals that a session has ended. Triggers a final memory extraction cycle for any unprocessed messages in that session.
Close the conversation: Signals that the entire conversation is complete across all sessions. Closing a conversation also closes any remaining open sessions.

Entry Categories

When ingesting interaction data, each entry belongs to one of three categories.

Category	Purpose
Message	Human-readable messages between participants. Each message requires content text and a participant identifier.
Trace	Agent reasoning steps, tool calls, and tool results. Traces capture what the agent did behind the scenes (not shown to end users). They include tool names, parameters, and results as structured metadata.
Entry	Generic category for other interaction data that doesn’t fit message or trace.

For message entries, provide the message content and identify the sender. For trace entries, provide structured metadata about the tool call or reasoning step. Optionally include timestamps, attachments, and interaction IDs that group related entries together—for example, a tool call and its result.

Key Ingestion Behaviors

When ingesting entries, these behaviors apply.

Ordering: Entries are processed in the order received. Include timestamps for accurate chronological sequencing.
Batching: Send multiple entries in a single API call for efficiency.
Fire-and-forget: The API response returns immediately after storage and indexing are committed, without waiting for memory extraction to complete.

Retrieve Conversations

The APIs support reading back conversation data. List conversations with filters (owner, status, time range, participant), retrieve a specific conversation with all its entries, sessions, and participants, or list sessions within a conversation.

Multi-Agent Session Patterns

The conversation APIs support bridge session handoff. This is a restricted context where Agent A closes its session and Agent B creates a new session within the same conversation. Agent B can access conversation-level context but not the details of Agent A’s session. Use this pattern when Agent A handled sensitive information.