openapi: 3.0.1 info: title: Pangea Security Services AI Guard API description: Specification of representative Pangea security service APIs. Pangea exposes each security capability as its own REST service reachable at https://{service}.{csp}.{geo}.pangea.cloud (for example https://redact.aws.us.pangea.cloud). All requests are POST with a JSON body and are authenticated with a Bearer service token (or OAuth 2 access token) in the Authorization header. This document models several representative services - AuthN, Secure Audit Log, Redact, Vault, File Scan, IP Intel, Domain Intel, and AI Guard - and is not exhaustive of every endpoint or field. termsOfService: https://pangea.cloud/legal/terms-of-service/ contact: name: Pangea Support url: https://pangea.cloud/contact/ version: '1.0' servers: - url: https://{service}.{csp}.{geo}.pangea.cloud description: Per-service Pangea host. Each service is reached at its own subdomain. variables: service: default: redact description: Service name (authn, audit, redact, vault, file-scan, ip-intel, domain-intel, ai-guard). csp: default: aws description: Cloud service provider hosting the service. geo: default: us description: Geographic region (us, eu). security: - bearerAuth: [] tags: - name: AI Guard description: Detect and redact malicious content in LLM inputs and outputs. paths: /v1/text/guard: servers: - url: https://ai-guard.{csp}.{geo}.pangea.cloud variables: csp: default: aws geo: default: us post: operationId: aiGuardText tags: - AI Guard summary: Guard LLM text. description: Detect, remove, or block malicious content and intent in LLM inputs and outputs to prevent model manipulation and data leakage. requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/AiGuardRequest' responses: '200': description: Guard result. content: application/json: schema: $ref: '#/components/schemas/AiGuardResponse' '401': $ref: '#/components/responses/Unauthorized' components: schemas: PangeaResponse: type: object description: Standard Pangea response envelope wrapping every service result. properties: request_id: type: string request_time: type: string format: date-time response_time: type: string format: date-time status: type: string example: Success summary: type: string result: type: object AiGuardResponse: allOf: - $ref: '#/components/schemas/PangeaResponse' - type: object properties: result: type: object properties: blocked: type: boolean transformed: type: boolean recipe: type: string detectors: type: object description: Per-detector analysis (prompt_injection, pii, secrets, malicious_entity). prompt_text: type: string AiGuardRequest: type: object properties: text: type: string description: Plain text input (up to 20 KiB). messages: type: array items: type: object properties: role: type: string content: type: string recipe: type: string default: pangea_prompt_guard debug: type: boolean default: false responses: Unauthorized: description: Missing or invalid authentication token. content: application/json: schema: $ref: '#/components/schemas/PangeaResponse' securitySchemes: bearerAuth: type: http scheme: bearer description: Pangea service token or OAuth 2 access token passed as a Bearer token.