AI conversation data type

A GovTribe AI conversation with project assignment, work status, generation settings, credit usage, and recall-aware access to visible messages.

Web: Manage conversations

Notes

Each conversation is an independent workstream. A project can group conversations, but it does not create a parent-child, manager-worker, or automatic continuation relationship between them.
GovTribe AI can manage only the customer's owned conversations in the same workspace. Hidden internal conversations are not included.
GovTribe AI can read visible messages from the current conversation. Reading another owned conversation's messages requires Full Conversation Recall; its metadata and working status remain available when recall is disabled.

The internal govtribe_type value for this data type is oai_response.

Attributes

Field	Type	Attribute
`govtribe_id`	`string`	GovTribe ID
`govtribe_type`	`string`	GovTribe type
`govtribe_url`	`string`	GovTribe URL
`name`	`string`	Name
`description`	`string`	Description
`conversation_excerpt`	`string \| null`	Conversation excerpt
`created_at`	`string \| null`	Created at
`updated_at`	`string \| null`	Updated at

Conversation management details

When GovTribe AI manages an owned conversation, the returned conversation record can include these additional details. These are management metadata, not extra fields in the search and recall representation above.

Returned detail	Meaning
`project`	The assigned project's ID and name, or `null` when the conversation is not assigned to a project. Project assignment can change without changing the conversation transcript.
`response_status`	The conversation's last persisted response status.
`is_working`	Whether the conversation currently has recent streaming activity. Use this as the current working indicator.
`is_streaming` and `streaming_started_at`	The persisted streaming flag and the time streaming started. These can outlive recent activity when a stream stalls, so they do not replace `is_working`.
`generation_settings`	The stored, selected, active-turn, and effective settings. Each available setting contains `model`, `reasoning`, and `verbosity`; stored or active-turn settings can be `null` when no value is saved or no turn is active.
`credit_usage`	GovTribe's canonical conversation credit summary, including formatted and exact credit totals and its update time. The value is `null` until a complete, trustworthy summary is available.
`messages`	A cursor-paginated page of visible user and assistant messages. The default page contains 20 messages and the maximum is 100. Tool output, setup messages, and hidden transcript content are excluded.
`messages_unavailable_reason`	`full_conversation_recall_disabled` when another conversation's metadata can be read but its visible messages cannot.

Inter-conversation message provenance

When GovTribe AI sends a message from one owned conversation to another, GovTribe attaches provenance from trusted conversation-tool context. The model cannot provide or override this provenance. The delivered content remains a user-role message; its attribution does not give it system or developer authority.

Visible transcript messages expose this safe projection:

{
  "provenance": {
    "type": "govtribe_ai_thread",
    "label": "Sent from another GovTribe AI thread",
    "source": {
      "id": "SOURCE_CONVERSATION_ID",
      "name": "Current source name",
      "url": "AUTHORIZED_URL",
      "showRoute": {}
    }
  }
}

source is null when the source conversation was deleted, is no longer accessible to the owner in the same workspace, or appears through a public share. The attribution label remains, but the unavailable source ID and link are not exposed. Direct customer messages do not include inter-conversation provenance.

Message delivery receipts

Create_AI_Conversation and message-bearing Update_AI_Conversation responses include an additive delivery_receipt beside the existing conversation fields and top-level operation:

{
  "delivery_receipt": {
    "target": {
      "id": "TARGET_CONVERSATION_ID",
      "name": "Current target name",
      "url": "AUTHORIZED_URL"
    },
    "message_id": "DELIVERED_MESSAGE_ID",
    "operation": {
      "type": "start",
      "result": {}
    }
  }
}

The operation type is start, steer, or recovery_queue, and operation.result preserves the exact lifecycle result, including replay, billing, recovery, queuing, or steering details. Inspect that result before claiming the destination started or accepted the message. A metadata-only or stop-only update has no delivery receipt.

Reuse prior AI work: Choose between projects, Full Conversation Recall, memories, and files when returning to earlier work.