LLMWhisperer Highlights API

openapi: 3.0.1
info:
  title: LLMWhisperer API
  description: >-
    LLMWhisperer (by Unstract / Zipstack) is a document-to-text extraction API
    that converts PDFs, scanned documents, and images into clean,
    layout-preserving text ready for large language models. The v2 API is
    asynchronous: submit a document to POST /whisper, poll GET /whisper-status,
    then fetch the result with GET /whisper-retrieve. GET /highlights returns
    per-line bounding-box coordinates and /whisper-manage-callback manages
    webhook callbacks. All requests authenticate with the unstract-key header.
  termsOfService: https://unstract.com/terms-of-service/
  contact:
    name: Unstract Support
    url: https://unstract.com/llmwhisperer/
  version: '2.0'
servers:
  - url: https://llmwhisperer-api.us-central.unstract.com/api/v2
    description: US Central region
  - url: https://llmwhisperer-api.eu-west.unstract.com/api/v2
    description: EU West region
security:
  - unstractKey: []
paths:
  /whisper:
    post:
      operationId: whisper
      tags:
        - Extraction
      summary: Submit a document for text extraction
      description: >-
        Converts a document to text. Accepts the raw document as binary
        (application/octet-stream) or, when url_in_post is true, a URL in the
        request body. Processing is asynchronous; a whisper_hash is returned to
        track and retrieve the job.
      parameters:
        - name: mode
          in: query
          description: Extraction mode.
          schema:
            type: string
            enum: [native_text, low_cost, high_quality, form, table]
            default: form
        - name: output_mode
          in: query
          description: Output formatting mode.
          schema:
            type: string
            enum: [layout_preserving, text]
            default: layout_preserving
        - name: page_seperator
          in: query
          description: Page delimiter string inserted between pages.
          schema:
            type: string
            default: <<<
        - name: pages_to_extract
          in: query
          description: 'Pages to extract, e.g. "1-5,7,21-".'
          schema:
            type: string
        - name: median_filter_size
          in: query
          description: Median filter size for low_cost mode noise removal.
          schema:
            type: integer
            default: 0
        - name: gaussian_blur_radius
          in: query
          description: Gaussian blur radius for low_cost mode noise removal.
          schema:
            type: number
            default: 0
        - name: line_splitter_tolerance
          in: query
          description: Baseline factor for line splitting (fraction of line height).
          schema:
            type: number
            default: 0.4
        - name: line_splitter_strategy
          in: query
          description: Line splitting strategy.
          schema:
            type: string
            default: left-priority
        - name: horizontal_stretch_factor
          in: query
          description: Horizontal stretch factor for multi-column layout adjustment.
          schema:
            type: number
            default: 1.0
        - name: url_in_post
          in: query
          description: When true, the request body is a document URL instead of binary data.
          schema:
            type: boolean
            default: false
        - name: mark_vertical_lines
          in: query
          description: Reproduce vertical layout lines in the output.
          schema:
            type: boolean
            default: false
        - name: mark_horizontal_lines
          in: query
          description: Reproduce horizontal layout lines in the output.
          schema:
            type: boolean
            default: false
        - name: lang
          in: query
          description: Language hint for OCR (ISO 639-2/B, e.g. eng).
          schema:
            type: string
            default: eng
        - name: tag
          in: query
          description: Auditing label associated with the request.
          schema:
            type: string
            default: default
        - name: file_name
          in: query
          description: Auditing reference file name.
          schema:
            type: string
        - name: use_webhook
          in: query
          description: Name of a registered webhook to deliver the result to.
          schema:
            type: string
        - name: webhook_metadata
          in: query
          description: Metadata echoed back to the webhook with the result.
          schema:
            type: string
        - name: add_line_nos
          in: query
          description: Enable line numbering and persist line metadata for highlights.
          schema:
            type: boolean
            default: false
        - name: allow_rotated_text
          in: query
          description: Include rotated/angled text in extraction.
          schema:
            type: boolean
            default: true
        - name: word_confidence_threshold
          in: query
          description: OCR word confidence filter (0-1).
          schema:
            type: number
            default: 0.3
      requestBody:
        required: true
        description: Document binary, or a document URL when url_in_post is true.
        content:
          application/octet-stream:
            schema:
              type: string
              format: binary
          text/plain:
            schema:
              type: string
      responses:
        '202':
          description: Whisper job accepted.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/WhisperAccepted'
        '400':
          description: Bad request.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/Error'
  /whisper-status:
    get:
      operationId: whisperStatus
      tags:
        - Status
      summary: Check the status of a whisper job
      parameters:
        - name: whisper_hash
          in: query
          required: true
          description: The whisper hash returned when the whisper job was started.
          schema:
            type: string
      responses:
        '200':
          description: Current status of the whisper job.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/WhisperStatus'
        '400':
          description: Whisper job not found for the provided whisper hash.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/Error'
  /whisper-retrieve:
    get:
      operationId: whisperRetrieve
      tags:
        - Retrieve
      summary: Retrieve the extracted text of a processed whisper job
      description: >-
        Returns the extracted text once the job is processed. For security and
        privacy, the extracted text can be retrieved only once.
      parameters:
        - name: whisper_hash
          in: query
          required: true
          description: The whisper hash returned when the whisper job was started.
          schema:
            type: string
        - name: text_only
          in: query
          description: When true, returns only the extracted text without metadata.
          schema:
            type: boolean
            default: false
      responses:
        '200':
          description: Extracted text and metadata.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/WhisperResult'
            text/plain:
              schema:
                type: string
        '400':
          description: Retrieval error.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/Error'
        '404':
          description: Invalid whisper hash.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/Error'
  /highlights:
    get:
      operationId: highlights
      tags:
        - Highlights
      summary: Retrieve per-line bounding-box coordinates
      description: >-
        Returns bounding-box metadata for the requested lines so callers can
        highlight extracted text in the source document. Requires the whisper
        job to have been submitted with add_line_nos enabled.
      parameters:
        - name: whisper_hash
          in: query
          required: true
          description: The whisper hash returned when the whisper job was started.
          schema:
            type: string
        - name: lines
          in: query
          required: true
          description: 'Lines to retrieve, e.g. "1-5,7,21-".'
          schema:
            type: string
      responses:
        '200':
          description: Map of line number to bounding-box metadata.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/HighlightsResponse'
        '400':
          description: Bad request.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/Error'
  /whisper-manage-callback:
    post:
      operationId: registerWebhook
      tags:
        - Webhooks
      summary: Register a webhook callback
      description: >-
        Registers a webhook the service calls with the extracted result when a
        document submitted with use_webhook finishes processing. A test payload
        with result_text "WEBHOOK_TEST" is sent on registration.
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/WebhookConfig'
      responses:
        '201':
          description: Webhook registered.
        '400':
          description: Registration failed (e.g. endpoint unreachable).
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/Error'
    get:
      operationId: getWebhook
      tags:
        - Webhooks
      summary: Retrieve a registered webhook
      parameters:
        - name: webhook_name
          in: query
          required: true
          schema:
            type: string
      responses:
        '200':
          description: Webhook details.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/WebhookConfig'
    put:
      operationId: updateWebhook
      tags:
        - Webhooks
      summary: Update a registered webhook
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/WebhookConfig'
      responses:
        '200':
          description: Webhook updated.
    delete:
      operationId: deleteWebhook
      tags:
        - Webhooks
      summary: Delete a registered webhook
      parameters:
        - name: webhook_name
          in: query
          required: true
          schema:
            type: string
      responses:
        '200':
          description: Webhook deleted.
components:
  securitySchemes:
    unstractKey:
      type: apiKey
      in: header
      name: unstract-key
      description: LLMWhisperer API key passed in the unstract-key request header.
  schemas:
    WhisperAccepted:
      type: object
      properties:
        message:
          type: string
          example: Whisper Job Accepted
        status:
          type: string
          example: processing
        whisper_hash:
          type: string
          example: xxxxx|xxx
    WhisperStatus:
      type: object
      properties:
        status:
          type: string
          enum: [accepted, processing, processed, error, retrieved]
        message:
          type: string
        detail:
          type: array
          items:
            type: object
            properties:
              page_no:
                type: integer
              message:
                type: string
              execution_time_in_seconds:
                type: number
    WhisperResult:
      type: object
      properties:
        result_text:
          type: string
          description: The extracted text from the document.
        confidence_metadata:
          type: array
          description: Per-line word confidence scores (words below 0.9 confidence).
          items:
            type: array
            items:
              type: object
        webhook_metadata:
          type: string
          description: Metadata sent to the webhook after the document is processed.
        metadata:
          type: object
          description: Reserved for future use.
    HighlightLine:
      type: object
      properties:
        base_y:
          type: integer
        base_y_percent:
          type: number
        height:
          type: integer
        height_percent:
          type: number
        page:
          type: integer
        page_height:
          type: integer
        raw:
          type: array
          description: '[page_no, y, height, page_height]'
          items:
            type: integer
    HighlightsResponse:
      type: object
      additionalProperties:
        $ref: '#/components/schemas/HighlightLine'
    WebhookConfig:
      type: object
      required:
        - url
        - webhook_name
      properties:
        url:
          type: string
          description: Destination URL the result is POSTed to.
        auth_token:
          type: string
          description: Optional bearer token sent to the webhook endpoint.
        webhook_name:
          type: string
          description: Identifier used to reference this webhook in use_webhook.
    Error:
      type: object
      properties:
        message:
          type: string