openapi: 3.0.1
info:
  title: PromptArmor API
  description: >-
    PromptArmor is an LLM application security platform that detects and blocks
    indirect prompt injection, data exfiltration, phishing, and system
    manipulation in production AI applications. Content submitted to the API is
    run through a detection engine of default and modifiable detectors and a
    fast verdict (for example containsInjection) is returned so the calling
    application can block or allow the content before a completion is acted on.


    NOTE: PromptArmor's full API reference at https://promptarmor.readme.io is
    gated behind authentication. This specification reflects the endpoints,
    authentication scheme (Api-Key header), base URL (https://api.promptarmor.com),
    and request/response fields that are publicly documented in PromptArmor's
    architecture overview and integration guides. Field names beyond the
    confirmed `containsInjection` verdict and the `content`, `text`,
    `session_id`, and `detectors` request parameters are modeled
    conservatively and should be reconciled against the authenticated docs.
  termsOfService: https://www.promptarmor.com/
  contact:
    name: PromptArmor
    url: https://www.promptarmor.com/
  version: '1.0'
servers:
  - url: https://api.promptarmor.com
    description: PromptArmor production API
security:
  - ApiKeyAuth: []
tags:
  - name: Content Check
    description: Single-call content verdict against the detection engine.
  - name: Analyze
    description: Analyze LLM input and output through the detector engine.
paths:
  /v1/check_content:
    post:
      operationId: checkContent
      tags:
        - Content Check
      summary: Check content for adversarial / injected instructions.
      description: >-
        Submits content (such as untrusted text an LLM is about to summarize or
        act on) to the PromptArmor detection engine and returns a verdict
        indicating whether the content matches known threat classes. The
        response includes a containsInjection boolean.
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/CheckContentRequest'
      responses:
        '200':
          description: Detection verdict.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/CheckContentResponse'
        '401':
          description: Missing or invalid Api-Key.
        '429':
          description: Rate limited.
  /v1/analyze/input:
    post:
      operationId: analyzeInput
      tags:
        - Analyze
      summary: Analyze LLM input.
      description: >-
        Analyzes LLM input (for example an email your LLM is about to
        summarize) and runs the detectors - including adversarial input
        detection - returning a verdict. When a session_id is supplied,
        session-based anomaly detection is also run.
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/AnalyzeRequest'
      responses:
        '200':
          description: Detection verdict.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/AnalyzeResponse'
        '401':
          description: Missing or invalid Api-Key.
        '429':
          description: Rate limited.
  /v1/analyze/output:
    post:
      operationId: analyzeOutput
      tags:
        - Analyze
      summary: Analyze LLM output.
      description: >-
        Analyzes LLM output (for example the summary your LLM produced) and runs
        the detectors - including data exfiltration detection - returning a
        verdict. When a session_id is supplied, session-based anomaly detection
        is also run.
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/AnalyzeRequest'
      responses:
        '200':
          description: Detection verdict.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/AnalyzeResponse'
        '401':
          description: Missing or invalid Api-Key.
        '429':
          description: Rate limited.
components:
  securitySchemes:
    ApiKeyAuth:
      type: apiKey
      in: header
      name: Api-Key
      description: PromptArmor API key passed in the Api-Key request header.
  schemas:
    CheckContentRequest:
      type: object
      required:
        - content
      properties:
        content:
          type: string
          description: The content to evaluate against the detection engine.
        session_id:
          type: string
          description: >-
            Optional session identifier. When supplied, PromptArmor also runs
            session-based anomaly detection across the session.
        detectors:
          type: array
          description: >-
            Optional list of detectors to apply in addition to / instead of the
            default detectors (data exfiltration, phishing, system
            manipulation, adversarial instructions).
          items:
            type: string
    CheckContentResponse:
      type: object
      properties:
        containsInjection:
          type: boolean
          description: >-
            True when the content matches a known injection / threat class. A
            detector returning true indicates an issue was found for that
            detection type.
    AnalyzeRequest:
      type: object
      required:
        - text
      properties:
        text:
          type: string
          description: >-
            The LLM input or output text to analyze (for example an email being
            summarized, or the generated summary).
        session_id:
          type: string
          description: >-
            Optional session identifier enabling session-based anomaly
            detection across multiple analyze calls.
        detectors:
          type: array
          description: Optional list of detectors to apply for this request.
          items:
            type: string
    AnalyzeResponse:
      type: object
      properties:
        containsInjection:
          type: boolean
          description: >-
            True when the analyzed text matches a known injection / threat
            class.
        detectors:
          type: object
          description: >-
            Per-detector results. A detector value of true indicates an issue
            was found for that detection type (for example data exfiltration,
            phishing, system manipulation, adversarial instructions).
          additionalProperties:
            type: boolean