Scale AI Task Retention API

openapi: 3.2.0
info:
  title: Agentex Task Retention API
  version: 0.1.0
tags:
- name: task-retention
paths:
  /tasks/{task_id}/export:
    get:
      tags:
      - task-retention
      summary: Export Task
      description: 'Build a self-contained snapshot of a task''s content surfaces.


        Returns the exact payload format that POST /rehydrate accepts, so

        export → clean → rehydrate is a round-trip-equivalent operation.'
      operationId: export_task_tasks__task_id__export_get
      parameters:
      - name: task_id
        in: path
        required: true
        schema:
          type: string
          title: Task Id
      responses:
        '200':
          description: Successful Response
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ExportTaskResponse'
        '422':
          description: Validation Error
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/HTTPValidationError'
    post:
      tags:
      - task-retention
      summary: Export Task To Url
      description: 'Build the task snapshot and PUT it to a caller-supplied presigned URL.


        Use this when the snapshot is too large for a JSON response body (long

        conversations, deep reasoning content, many attachments). The upload URL

        must be https and resolve to a public address — see SSRF guard.'
      operationId: export_task_to_url_tasks__task_id__export_post
      parameters:
      - name: task_id
        in: path
        required: true
        schema:
          type: string
          title: Task Id
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/ExportTaskToUrlRequest'
      responses:
        '200':
          description: Successful Response
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ExportTaskToUrlResponse'
        '422':
          description: Validation Error
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/HTTPValidationError'
  /tasks/{task_id}/clean:
    post:
      tags:
      - task-retention
      summary: Clean Task
      description: 'Delete content-bearing rows for a stale task.


        Refuses on active tasks, in-flight workflows, or unprocessed events

        regardless of `force`. The `force=true` flag only bypasses the

        idle-threshold check.'
      operationId: clean_task_tasks__task_id__clean_post
      parameters:
      - name: task_id
        in: path
        required: true
        schema:
          type: string
          title: Task Id
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/CleanTaskRequest'
      responses:
        '200':
          description: Successful Response
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/CleanTaskResponse'
        '422':
          description: Validation Error
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/HTTPValidationError'
  /tasks/{task_id}/rehydrate:
    post:
      tags:
      - task-retention
      summary: Rehydrate Task
      description: 'Restore content-bearing rows from a snapshot.


        Two modes:

        - Inline: caller provides messages and task_states in the request body.

        - URL: caller provides snapshot_url; Agentex downloads and parses it.


        Refuses if the task isn''t currently in a cleaned state, or if any supplied

        message/state ID already exists in Mongo (catches double-rehydrate).'
      operationId: rehydrate_task_tasks__task_id__rehydrate_post
      parameters:
      - name: task_id
        in: path
        required: true
        schema:
          type: string
          title: Task Id
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/RehydrateTaskRequest'
      responses:
        '204':
          description: Successful Response
        '422':
          description: Validation Error
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/HTTPValidationError'
components:
  schemas:
    RehydrateTaskRequest:
      properties:
        task_id:
          type: string
          title: Task Id
        messages:
          items:
            $ref: '#/components/schemas/TaskMessageEntity'
          type: array
          title: Messages
        task_states:
          items:
            $ref: '#/components/schemas/StateEntity'
          type: array
          title: Task States
        snapshot_url:
          anyOf:
          - type: string
            maxLength: 2083
            minLength: 1
            format: uri
          - type: 'null'
          title: Snapshot Url
          description: Presigned GET URL whose body is a JSON-encoded TaskSnapshotEntity. Must be https; must resolve to a public address. When set, messages/task_states must be empty.
      type: object
      required:
      - task_id
      title: RehydrateTaskRequest
      description: 'Either provide inline content (messages + task_states) or a snapshot_url

        pointing at a presigned JSON download. Mixing both is rejected.


        The inline form is the canonical shape used by export''s GET response, so

        snapshot → clean → rehydrate round-trips cleanly without serialization

        changes.'
    ToolResponseContentEntity:
      properties:
        type:
          type: string
          const: tool_response
          title: Type
          description: The type of the message, in this case `tool_response`.
          default: tool_response
        author:
          $ref: '#/components/schemas/MessageAuthor'
          description: The role of the messages author, in this case `system`, `user`, `assistant`, or `tool`.
        style:
          $ref: '#/components/schemas/MessageStyle'
          description: The style of the message. This is used by the client to determine how to display the message.
          default: static
        tool_call_id:
          type: string
          title: Tool Call Id
          description: The ID of the tool call that is being responded to.
        name:
          type: string
          title: Name
          description: The name of the tool that is being responded to.
        content:
          title: Content
          description: The result of the tool.
        is_error:
          anyOf:
          - type: boolean
          - type: 'null'
          title: Is Error
          description: Whether the tool call resulted in an error. `None` when the harness does not report a status.
      type: object
      required:
      - author
      - tool_call_id
      - name
      - content
      title: ToolResponseContentEntity
    TaskMessageEntity:
      properties:
        id:
          anyOf:
          - type: string
          - type: 'null'
          title: Id
          description: The task message's unique id
        task_id:
          type: string
          title: Task Id
          description: ID of the task this message belongs to
        content:
          oneOf:
          - $ref: '#/components/schemas/TextContentEntity'
          - $ref: '#/components/schemas/DataContentEntity'
          - $ref: '#/components/schemas/ToolRequestContentEntity'
          - $ref: '#/components/schemas/ToolResponseContentEntity'
          - $ref: '#/components/schemas/ReasoningContentEntity'
          title: Content
          description: The content of the message. This content is not OpenAI compatible. These are messages that are meant to be displayed to the user.
          discriminator:
            propertyName: type
            mapping:
              data: '#/components/schemas/DataContentEntity'
              reasoning: '#/components/schemas/ReasoningContentEntity'
              text: '#/components/schemas/TextContentEntity'
              tool_request: '#/components/schemas/ToolRequestContentEntity'
              tool_response: '#/components/schemas/ToolResponseContentEntity'
        streaming_status:
          anyOf:
          - type: string
            enum:
            - IN_PROGRESS
            - DONE
          - type: 'null'
          title: In case of streaming, this indicates whether the message is still being streamed or has been completed
        created_at:
          anyOf:
          - type: string
            format: date-time
          - type: 'null'
          title: Created At
          description: The timestamp when the message was created
        updated_at:
          anyOf:
          - type: string
            format: date-time
          - type: 'null'
          title: Updated At
          description: The timestamp when the message was last updated
      type: object
      required:
      - task_id
      - content
      title: TaskMessageEntity
      description: 'Represents a message in the agent system.


        This entity is used to store messages in MongoDB, with each message

        associated with a specific task.'
    TextContentEntity:
      properties:
        type:
          type: string
          const: text
          title: Type
          description: The type of the message, in this case `text`.
          default: text
        author:
          $ref: '#/components/schemas/MessageAuthor'
          description: The role of the messages author, in this case `system`, `user`, `assistant`, or `tool`.
        style:
          $ref: '#/components/schemas/MessageStyle'
          description: The style of the message. This is used by the client to determine how to display the message.
          default: static
        format:
          $ref: '#/components/schemas/TextFormat'
          description: The format of the message. This is used by the client to determine how to display the message.
          default: plain
        content:
          type: string
          title: Content
          description: The contents of the text message.
        attachments:
          anyOf:
          - items:
              $ref: '#/components/schemas/FileAttachmentEntity'
            type: array
          - type: 'null'
          title: Attachments
          description: Optional list of file attachments with structured metadata.
      type: object
      required:
      - author
      - content
      title: TextContentEntity
    StateEntity:
      properties:
        id:
          anyOf:
          - type: string
          - type: 'null'
          title: Id
          description: The task state's unique id
        task_id:
          type: string
          title: Task Id
          description: ID of the task this state belongs to. The combination of task_id and agent_id is globally unique.
        agent_id:
          type: string
          title: Agent Id
          description: ID of the agent this state belongs to. The combination of task_id and agent_id is globally unique.
        state:
          additionalProperties: true
          type: object
          title: State
          description: The state object that contains arbitrary data
        created_at:
          anyOf:
          - type: string
            format: date-time
          - type: 'null'
          title: Created At
          description: The timestamp when the state was created
        updated_at:
          anyOf:
          - type: string
            format: date-time
          - type: 'null'
          title: Updated At
          description: The timestamp when the state was last updated
      type: object
      required:
      - task_id
      - agent_id
      - state
      title: StateEntity
      description: 'Represents a state in the agent system. A state is associated uniquely with a task and an agent.


        This entity is used to store states in MongoDB, with each state

        associated with a specific task and agent. The combination of task_id and agent_id is globally unique.


        The state is a dictionary of arbitrary data.'
    ExportTaskResponse:
      properties:
        task_id:
          type: string
          title: Task Id
        messages:
          items:
            $ref: '#/components/schemas/TaskMessageEntity'
          type: array
          title: Messages
        task_states:
          items:
            $ref: '#/components/schemas/StateEntity'
          type: array
          title: Task States
      type: object
      required:
      - task_id
      title: ExportTaskResponse
      description: Wire format mirrors the entity directly — schema parity is intentional.
    HTTPValidationError:
      properties:
        detail:
          items:
            $ref: '#/components/schemas/ValidationError'
          type: array
          title: Detail
      type: object
      title: HTTPValidationError
    ReasoningContentEntity:
      properties:
        type:
          type: string
          const: reasoning
          title: Type
          description: The type of the message, in this case `reasoning`.
          default: reasoning
        author:
          $ref: '#/components/schemas/MessageAuthor'
          description: The role of the messages author, in this case `system`, `user`, `assistant`, or `tool`.
        style:
          $ref: '#/components/schemas/MessageStyle'
          description: The style of the message. This is used by the client to determine how to display the message.
          default: static
        summary:
          items:
            type: string
          type: array
          title: Summary
          description: A list of short reasoning summaries
        content:
          anyOf:
          - items:
              type: string
            type: array
          - type: 'null'
          title: Content
          description: The reasoning content or chain-of-thought text
      type: object
      required:
      - author
      - summary
      title: ReasoningContentEntity
    DataContentEntity:
      properties:
        type:
          type: string
          const: data
          title: Type
          description: The type of the message, in this case `data`.
          default: data
        author:
          $ref: '#/components/schemas/MessageAuthor'
          description: The role of the messages author, in this case `system`, `user`, `assistant`, or `tool`.
        style:
          $ref: '#/components/schemas/MessageStyle'
          description: The style of the message. This is used by the client to determine how to display the message.
          default: static
        data:
          additionalProperties: true
          type: object
          title: Data
          description: The contents of the data message.
      type: object
      required:
      - author
      - data
      title: DataContentEntity
    ExportTaskToUrlResponse:
      properties:
        task_id:
          type: string
          title: Task Id
        upload_url:
          type: string
          title: Upload Url
        uploaded_bytes:
          type: integer
          title: Uploaded Bytes
        messages_count:
          type: integer
          title: Messages Count
        task_states_count:
          type: integer
          title: Task States Count
      type: object
      required:
      - task_id
      - upload_url
      - uploaded_bytes
      - messages_count
      - task_states_count
      title: ExportTaskToUrlResponse
    CleanTaskResponse:
      properties:
        task_id:
          type: string
          title: Task Id
        cleaned_at:
          type: string
          format: date-time
          title: Cleaned At
        messages_deleted:
          type: integer
          title: Messages Deleted
        task_states_deleted:
          type: integer
          title: Task States Deleted
        events_deleted:
          type: integer
          title: Events Deleted
      type: object
      required:
      - task_id
      - cleaned_at
      - messages_deleted
      - task_states_deleted
      - events_deleted
      title: CleanTaskResponse
    FileAttachmentEntity:
      properties:
        file_id:
          type: string
          title: File Id
          description: The unique ID of the attached file
        name:
          type: string
          title: Name
          description: The name of the file
        size:
          type: integer
          title: Size
          description: The size of the file in bytes
        type:
          type: string
          title: Type
          description: The MIME type or content type of the file
      type: object
      required:
      - file_id
      - name
      - size
      - type
      title: FileAttachmentEntity
      description: Represents a file attachment in messages.
    ExportTaskToUrlRequest:
      properties:
        upload_url:
          type: string
          maxLength: 2083
          minLength: 1
          format: uri
          title: Upload Url
          description: Presigned PUT URL where Agentex will upload the task snapshot as JSON. Must be https; must resolve to a public address.
      type: object
      required:
      - upload_url
      title: ExportTaskToUrlRequest
    MessageStyle:
      type: string
      enum:
      - static
      - active
      title: MessageStyle
    TextFormat:
      type: string
      enum:
      - markdown
      - plain
      - code
      title: TextFormat
    ToolRequestContentEntity:
      properties:
        type:
          type: string
          const: tool_request
          title: Type
          description: The type of the message, in this case `tool_request`.
          default: tool_request
        author:
          $ref: '#/components/schemas/MessageAuthor'
          description: The role of the messages author, in this case `system`, `user`, `assistant`, or `tool`.
        style:
          $ref: '#/components/schemas/MessageStyle'
          description: The style of the message. This is used by the client to determine how to display the message.
          default: static
        tool_call_id:
          type: string
          title: Tool Call Id
          description: The ID of the tool call that is being requested.
        name:
          type: string
          title: Name
          description: The name of the tool that is being requested.
        arguments:
          additionalProperties: true
          type: object
          title: Arguments
          description: The arguments to the tool.
      type: object
      required:
      - author
      - tool_call_id
      - name
      - arguments
      title: ToolRequestContentEntity
    ValidationError:
      properties:
        loc:
          items:
            anyOf:
            - type: string
            - type: integer
          type: array
          title: Location
        msg:
          type: string
          title: Message
        type:
          type: string
          title: Error Type
        input:
          title: Input
        ctx:
          type: object
          title: Context
      type: object
      required:
      - loc
      - msg
      - type
      title: ValidationError
    CleanTaskRequest:
      properties:
        force:
          type: boolean
          title: Force
          description: Skip the idle-threshold check. Active-workflow and unprocessed-events checks still apply. Admin use only.
          default: false
        idle_days:
          type: integer
          minimum: 1.0
          title: Idle Days
          description: Idle threshold in days (ignored when force=true).
          default: 7
      type: object
      title: CleanTaskRequest
    MessageAuthor:
      type: string
      enum:
      - user
      - agent
      title: MessageAuthor