openapi: 3.1.0
info:
  title: Compresr Platform Admin Compression API
  version: 1.0.0
tags:
- name: Compression
paths:
  /api/compress/question-specific/:
    post:
      tags:
      - Compression
      summary: Compress Query Specific
      description: 'Compress single context with question-aware filtering.


        For multiple context+query pairs, use the /batch endpoint.'
      operationId: compress_query_specific_api_compress_question_specific__post
      parameters:
      - name: X-API-Key
        in: header
        required: true
        schema:
          type: string
          title: X-Api-Key
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/QuerySpecificCompressionRequest'
      responses:
        '200':
          description: Successful Response
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/QuerySpecificCompressionResponse'
        '422':
          description: Validation Error
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/HTTPValidationError'
  /api/compress/question-specific/batch:
    post:
      tags:
      - Compression
      summary: Compress Query Specific Batch
      description: Batch compress multiple contexts with queries.
      operationId: compress_query_specific_batch_api_compress_question_specific_batch_post
      parameters:
      - name: X-API-Key
        in: header
        required: true
        schema:
          type: string
          title: X-Api-Key
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/QuerySpecificCompressionBatchRequest'
      responses:
        '200':
          description: Successful Response
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/QuerySpecificCompressionBatchResponse'
        '422':
          description: Validation Error
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/HTTPValidationError'
  /api/compress/question-specific/stream:
    post:
      tags:
      - Compression
      summary: Compress Query Specific Stream
      description: Stream question-specific compression result over SSE.
      operationId: compress_query_specific_stream_api_compress_question_specific_stream_post
      parameters:
      - name: X-API-Key
        in: header
        required: true
        schema:
          type: string
          title: X-Api-Key
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/QuerySpecificCompressionRequest'
      responses:
        '200':
          description: Successful Response
          content:
            application/json:
              schema: {}
        '422':
          description: Validation Error
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/HTTPValidationError'
  /api/compress/question-specific/health:
    get:
      tags:
      - Compression
      summary: Health Query Specific
      description: Health check for question-specific compression.
      operationId: health_query_specific_api_compress_question_specific_health_get
      responses:
        '200':
          description: Successful Response
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/HealthResponse'
  /api/compress/question-specific/models:
    get:
      tags:
      - Compression
      summary: Models Query Specific
      description: List available question-specific compression models (admin-only hidden for non-admins).
      operationId: models_query_specific_api_compress_question_specific_models_get
      security:
      - HTTPBearer: []
      parameters:
      - name: X-API-Key
        in: header
        required: false
        schema:
          anyOf:
          - type: string
          - type: 'null'
          title: X-Api-Key
      responses:
        '200':
          description: Successful Response
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/DemoDataResponse'
        '422':
          description: Validation Error
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/HTTPValidationError'
  /api/compress/question-specific/models/{model_id}:
    get:
      tags:
      - Compression
      summary: Model Query Specific
      description: Get details for a specific question-specific compression model.
      operationId: model_query_specific_api_compress_question_specific_models__model_id__get
      security:
      - HTTPBearer: []
      parameters:
      - name: model_id
        in: path
        required: true
        schema:
          type: string
          title: Model Id
      - name: X-API-Key
        in: header
        required: false
        schema:
          anyOf:
          - type: string
          - type: 'null'
          title: X-Api-Key
      responses:
        '200':
          description: Successful Response
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/DemoDataResponse'
        '422':
          description: Validation Error
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/HTTPValidationError'
components:
  schemas:
    CompressionResult:
      properties:
        original_tokens:
          type: integer
          title: Original Tokens
          description: Token count of original context
        compressed_tokens:
          type: integer
          title: Compressed Tokens
          description: Token count after compression
        target_compression_ratio:
          anyOf:
          - type: number
          - type: 'null'
          title: Target Compression Ratio
          description: User-requested compression ratio (if provided)
        actual_compression_ratio:
          type: number
          title: Actual Compression Ratio
          description: Actual achieved compression ratio (0-1)
        tokens_saved:
          type: integer
          title: Tokens Saved
          description: Number of tokens saved
        duration_ms:
          type: integer
          title: Duration Ms
          description: Processing time in milliseconds
        compressed_context:
          type: string
          title: Compressed Context
          description: Compressed context text
      type: object
      required:
      - original_tokens
      - compressed_tokens
      - actual_compression_ratio
      - tokens_saved
      - duration_ms
      - compressed_context
      title: CompressionResult
      description: 'Canonical single-context compression result.


        Used by query-specific routes, demo, and agentic responses.'
    SampleText:
      properties:
        name:
          type: string
          title: Name
        text:
          type: string
          title: Text
      type: object
      required:
      - name
      - text
      title: SampleText
    Source:
      type: string
      enum:
      - demo
      - extension
      - sdk:python
      - sdk:typescript
      - sdk:curl
      - gateway:unknown
      - gateway:anthropic
      - gateway:openai
      - gateway:gemini
      - integration:litellm
      - integration:hermes
      title: Source
      description: 'Source of an API request.


        Format: source[:detail]

        - demo, extension: no detail needed

        - sdk:python, sdk:typescript, sdk:curl

        - gateway:anthropic, gateway:openai, gateway:gemini

        - integration:litellm, integration:hermes'
    HealthResponse:
      properties:
        status:
          type: string
          title: Status
          description: Service status
        compression_server:
          type: boolean
          title: Compression Server
          description: Whether compression server is available
      type: object
      required:
      - status
      - compression_server
      title: HealthResponse
    QuerySpecificCompressionBatchResult:
      properties:
        total_original_tokens:
          type: integer
          title: Total Original Tokens
          default: 0
        total_compressed_tokens:
          type: integer
          title: Total Compressed Tokens
          default: 0
        total_tokens_saved:
          type: integer
          title: Total Tokens Saved
          default: 0
        average_compression_ratio:
          type: number
          title: Average Compression Ratio
          default: 0.0
        count:
          type: integer
          title: Count
          default: 0
        results:
          items:
            $ref: '#/components/schemas/CompressionResult'
          type: array
          title: Results
      type: object
      title: QuerySpecificCompressionBatchResult
    QuerySpecificCompressionResponse:
      properties:
        success:
          type: boolean
          title: Success
          default: true
        message:
          anyOf:
          - type: string
          - type: 'null'
          title: Message
        data:
          anyOf:
          - $ref: '#/components/schemas/CompressionResult'
          - type: 'null'
      type: object
      title: QuerySpecificCompressionResponse
    DemoCompressionModelOption:
      properties:
        value:
          type: string
          title: Value
        label:
          type: string
          title: Label
        description:
          type: string
          title: Description
          default: ''
        tier_scale:
          anyOf:
          - type: string
          - type: 'null'
          title: Tier Scale
      type: object
      required:
      - value
      - label
      title: DemoCompressionModelOption
    ValidationError:
      properties:
        loc:
          items:
            anyOf:
            - type: string
            - type: integer
          type: array
          title: Location
        msg:
          type: string
          title: Message
        type:
          type: string
          title: Error Type
      type: object
      required:
      - loc
      - msg
      - type
      title: ValidationError
    DemoDataResponse:
      properties:
        success:
          type: boolean
          title: Success
          default: true
        message:
          anyOf:
          - type: string
          - type: 'null'
          title: Message
        data:
          $ref: '#/components/schemas/DemoData'
      type: object
      required:
      - data
      title: DemoDataResponse
    QuerySpecificCompressionRequest:
      properties:
        context:
          anyOf:
          - type: string
          - type: 'null'
          title: Context
          description: Context text to compress (single string). If null or empty, the endpoint returns an empty compressed_context with zero tokens and no billing — no error is raised.
        compression_model_name:
          type: string
          title: Compression Model Name
          description: Compression model to use (e.g., 'latte_v1')
        query:
          type: string
          minLength: 1
          title: Query
          description: Query for context-aware filtering (REQUIRED)
        target_compression_ratio:
          anyOf:
          - type: number
            maximum: 200.0
            minimum: 0.0
          - type: 'null'
          title: Target Compression Ratio
          description: 'Target ratio: 0-1 (strength) or >1 for factor (e.g., 2=2x). Max 200.'
        coarse:
          type: boolean
          title: Coarse
          description: 'Use coarse-grained compression (paragraph-level). Default: True.'
          default: true
        heuristic_chunking:
          type: boolean
          title: Heuristic Chunking
          description: 'Use heuristic chunking for better structure preservation. Default: False.'
          default: false
        disable_placeholders:
          type: boolean
          title: Disable Placeholders
          description: 'Disable placeholder tokens in compressed output. Default: False.'
          default: false
        dynamic:
          type: boolean
          title: Dynamic
          description: latte_v2 only. Use Kneedle elbow selection instead of a fixed ratio; overrides target_compression_ratio when True.
          default: false
        dynamic_min_ratio:
          anyOf:
          - type: number
            minimum: 1.0
          - type: 'null'
          title: Dynamic Min Ratio
          description: latte_v2 only. Floor on compression when dynamic=True.
        dynamic_max_ratio:
          anyOf:
          - type: number
            minimum: 1.0
          - type: 'null'
          title: Dynamic Max Ratio
          description: latte_v2 only. Ceiling on compression when dynamic=True.
        source:
          anyOf:
          - $ref: '#/components/schemas/Source'
          - type: 'null'
          description: 'Source of request: demo, extension, sdk:<client>, gateway:<provider>'
      type: object
      required:
      - compression_model_name
      - query
      title: QuerySpecificCompressionRequest
      description: 'Request for single question-specific compression.


        Use /batch endpoint for multiple context+query pairs.

        Backbone: GemFilter'
    QuerySpecificCompressionBatchRequest:
      properties:
        inputs:
          items:
            $ref: '#/components/schemas/QuerySpecificCompressionBatchInput'
          type: array
          maxItems: 100
          minItems: 1
          title: Inputs
          description: List of inputs to compress
        compression_model_name:
          type: string
          title: Compression Model Name
          description: Compression model to use
        target_compression_ratio:
          anyOf:
          - type: number
            maximum: 200.0
            minimum: 0.0
          - type: 'null'
          title: Target Compression Ratio
          description: 'Target ratio: 0-1 (strength) or >1 for factor. Max 200.'
        coarse:
          type: boolean
          title: Coarse
          description: 'Use coarse-grained compression (paragraph-level). Applies to all items. Default: True.'
          default: true
        dynamic:
          type: boolean
          title: Dynamic
          description: latte_v2 only. Use Kneedle elbow selection instead of a fixed ratio; overrides target_compression_ratio when True.
          default: false
        dynamic_min_ratio:
          anyOf:
          - type: number
            minimum: 1.0
          - type: 'null'
          title: Dynamic Min Ratio
          description: latte_v2 only. Floor on compression when dynamic=True.
        dynamic_max_ratio:
          anyOf:
          - type: number
            minimum: 1.0
          - type: 'null'
          title: Dynamic Max Ratio
          description: latte_v2 only. Ceiling on compression when dynamic=True.
        source:
          anyOf:
          - $ref: '#/components/schemas/Source'
          - type: 'null'
          description: 'Source of request: demo, extension, sdk:<client>, gateway:<provider>'
      type: object
      required:
      - inputs
      - compression_model_name
      title: QuerySpecificCompressionBatchRequest
      description: 'Batch request for question-specific compression (per-item context+query pairs).


        Each input has its own context and query. Use this when you need different

        queries for different contexts.'
    QuerySpecificCompressionBatchInput:
      properties:
        context:
          anyOf:
          - type: string
          - type: 'null'
          title: Context
          description: Context text to compress. If null or empty, the corresponding result has compressed_context='' with zero tokens and is not billed.
        query:
          type: string
          minLength: 1
          title: Query
          description: Query for this context (REQUIRED)
      type: object
      required:
      - query
      title: QuerySpecificCompressionBatchInput
    DemoData:
      properties:
        sample_texts:
          items:
            $ref: '#/components/schemas/SampleText'
          type: array
          title: Sample Texts
        compression_models:
          items:
            $ref: '#/components/schemas/DemoCompressionModelOption'
          type: array
          title: Compression Models
          default: []
      type: object
      required:
      - sample_texts
      title: DemoData
    QuerySpecificCompressionBatchResponse:
      properties:
        success:
          type: boolean
          title: Success
          default: true
        message:
          anyOf:
          - type: string
          - type: 'null'
          title: Message
        data:
          anyOf:
          - $ref: '#/components/schemas/QuerySpecificCompressionBatchResult'
          - type: 'null'
      type: object
      title: QuerySpecificCompressionBatchResponse
    HTTPValidationError:
      properties:
        detail:
          items:
            $ref: '#/components/schemas/ValidationError'
          type: array
          title: Detail
      type: object
      title: HTTPValidationError
  securitySchemes:
    HTTPBearer:
      type: http
      scheme: bearer