Sign in
apex

Processes the scan and police request.

post/api/v1/police

Request body

anonymization'FixedSize' | 'VariableSize'

How to anonymize the data. If deanonymize is true, then VariablSize is required.

conversationIDstring

Identifies the conversation this request belongs to. Apex records it on the resulting log, which is what groups the successive requests of one conversation together in the logs and in the conversation view, and it is also made available to the analyzers and to policies. Send the same value on every request of the same conversation.

direction'Egress' | 'Ingress'

The direction of the traffic for this request, relative to the app component the caller's token identifies. Determines whether the ingress or the egress policies of that app component are evaluated.

messagesstring[]

Messages to process and provide detections for. Use data in extractions for processing binary data.

providerstring

The name of the provider to use for policy resolutions. Must not be set when destination app and component are set. On the police API an egress request must set either this or the destination app and component. On the scan API both may be omitted to run a plain scan that targets nothing.

toolsobject

The various tools used by the request.

type'Input' | 'Output'

The type of text.

Example request

{
  "conversationID": "6f37d752-bce1-4973-88f6-28b6c100ceb8",
  "destination": {
    "app": "other-ai-app",
    "component": "backend",
    "host": "api.openai.com",
    "ip": "192.0.2.42",
    "port": 443
  },
  "extractions": [
    {
      "toolResults": [
        {
          "callID": "toolu_019X5QaEeVTDFrQPHqMMgd1n",
          "name": "fetch",
          "serverName": "my-mcp-server"
        }
      ],
      "toolUses": [
        {
          "callID": "toolu_019X5QaEeVTDFrQPHqMMgd1n",
          "name": "get_weather",
          "serverName": "deepwiki"
        }
      ]
    }
  ],
  "messages": [
    "Summarize the main points of this article in bullet points.",
    "Generate a list of creative product names for a futuristic tech gadget."
  ],
  "provider": "openai",
  "source": {
    "ip": "192.0.2.42",
    "userClaims": [
      "name=John Doe",
      "email=john.doe@acme.com",
      "@validated=false"
    ],
    "username": "john.doe@acme.com"
  },
  "trace": {
    "parentSpanID": "00f067aa0ba902b7",
    "traceID": "4bf92f3577b34da6a3ce929d0e0e4736"
  }
}

Response

The request was processed and the response carries the enforcement decision. A platform failure inside the pipeline is also reported here rather than as an HTTP error: the response then carries the decision the application's failClose strategy dictates, Deny when failing closed and Allow when failing open, together with the structured error and the decision that was intended.

IDstring

ID is the identifier of the object.

annotationsobject

Annotations attached to the log.

clientstring

The client used to send the request.

clientVersionstring

The version of the client used to send the request.

contentRedactedboolean

If true, the content of the extractions was stripped from the audit entry for this request, and only the analysis and other metadata were kept. This is driven by the policy that produced this decision, so it is reported here because the caller has no other way to know it happened.

decision'Deny' | 'Allow' | 'Ask' | 'Report' | 'Bypassed' | 'ForbiddenUser' | 'Skipped' | 'Redirected' | 'NotApplicable' | 'Error' | 'UpstreamError'

User-facing outcome of the roundtrip. Reflects the policy engine's verdict, or in case of platform failure, the result of the failClose strategy (Deny on fail-close, Allow on fail-open, with structured error field carring the detail). NotApplicable is used by the scan and police APIs, which analyze content without rendering an enforcement decision. Error and UpstreamError stay in the allowed_choices list for backward compatibility with clients that still PUT those values; new round-trips never emit them — platform/upstream failures now surface via the structured Error field instead. NOTE: safe to drop Error and UpstreamError from this enum on or after 2026-07-19 (two months after the structured RoundtripError landed on 2026-05-19), once consumers have rolled forward.

hashstring

The hash of the input.

modelstring

The model used by the request.

namespacestring

The namespace of the object.

offbandboolean

If true, the policy that produced this decision asked for the analysis to run offband. The decision was therefore made without waiting for the analyzers, so the extractions in this response carry no detections, and any redaction the analyzers would have found was not applied. The full analyzer set runs after this response is sent, so the stored roundtrip for this request can report a stricter outcome than the one reported here.

permissiveboolean

If true, the policy that produced this decision is configured in permissive mode, so the content decision reported here is what the policy would have enforced and is not meant to be enforced. A caller acting on this response must let the request through when the decision is Deny, Ask or Report. This covers the content decision only. A decision of ForbiddenUser comes from the access policy, which permissive does not affect, and must still be enforced.

pipelineNamestring

The name of the particular pipeline that extracted the text.

providerstring

The provider to use.

providerType'LLM' | 'MCPServer' required

The type of the provider.

reasonsstring[]

The various reasons returned by the policy engine.

timestring date-time

Set the time of the message request.

toolsobject

The various tools used by the request.

type'Input' | 'Output'

The type of text.

Example response

{
  "alerts": [
    {
      "alertDefinition": "warning-notification",
      "principal": {
        "IP": "192.0.2.42",
        "app": {
          "component": "frontend",
          "labels": [
            "country=us",
            "another-label"
          ],
          "name": "MyApp",
          "userClaims": [
            "name=John Doe",
            "email=john.doe@acme.com",
            "@validated=false"
          ],
          "username": "john.doe@acme.com",
          "workloadGroupHash": "wg-0ff92a76a3765740e26d84947d92e5fc",
          "workloadGroupLabel": "k8s:deployment=mcp-chatbot-agent,namespace=demo",
          "workloadGroupSetHash": "wgs-0ff92a76a3765740e26d84947d92e5fc",
          "workloadGroupSetLabel": "k8s:namespace=demo"
        },
        "external": {
          "userClaims": [
            "name=John Doe",
            "email=john.doe@acme.com",
            "@validated=false"
          ],
          "username": "john.doe@acme.com",
          "workloadGroupHash": "wg-0ff92a76a3765740e26d84947d92e5fc",
          "workloadGroupLabel": "k8s:deployment=mcp-chatbot-agent,namespace=demo",
          "workloadGroupSetHash": "wgs-0ff92a76a3765740e26d84947d92e5fc",
          "workloadGroupSetLabel": "k8s:namespace=demo"
        },
        "tokenID": "1234-1224-123-1",
        "tokenName": "my-user-token",
        "user": {
          "name": "user@company.com"
        }
      }
    }
  ],
  "client": "curl",
  "clientVersion": "7.64.1",
  "destination": {
    "app": "MyApp",
    "component": "frontend",
    "host": "api.openai.com",
    "ip": "192.0.2.42",
    "labels": [
      "country=us",
      "another-label"
    ],
    "workloadGroupHash": "wg-0ff92a76a3765740e26d84947d92e5fc",
    "workloadGroupLabel": "k8s:deployment=mcp-chatbot-agent,namespace=demo",
    "workloadGroupSetHash": "wgs-0ff92a76a3765740e26d84947d92e5fc",
    "workloadGroupSetLabel": "k8s:namespace=demo"
  },
  "extractions": [
    {
      "PIIs": {
        "ssn": 0.8
      },
      "categories": [
        {
          "group": "image",
          "type": "png"
        }
      ],
      "confidentiality": 0.9,
      "customDataTypes": {
        "my_cdt": 1
      },
      "dataSets": {
        "cds": {
          "ct1": 1,
          "ct2": 2
        }
      },
      "exploits": {
        "prompt_injection": 0.8
      },
      "intent": {
        "write": 0.8
      },
      "keywords": {
        "my_keywork": 0.8
      },
      "languages": {
        "english": 0.8
      },
      "malcontents": {
        "toxic": 0.8
      },
      "modalities": [
        {
          "group": "image",
          "type": "png"
        }
      ],
      "relevance": 0.9,
      "secrets": {
        "credentials": 0.7
      },
      "toolResults": [
        {
          "callID": "toolu_019X5QaEeVTDFrQPHqMMgd1n",
          "name": "fetch",
          "serverName": "my-mcp-server"
        }
      ],
      "toolUses": [
        {
          "callID": "toolu_019X5QaEeVTDFrQPHqMMgd1n",
          "name": "get_weather",
          "serverName": "deepwiki"
        }
      ],
      "topics": {
        "category/enterprise": 0.7,
        "department/logistics": 0.8,
        "depict/document": 0.8,
        "extracted/typed_text_content": 1,
        "timeframe/current_year": 0.6
      }
    }
  ],
  "mcpMessage": {
    "direction": "Client2Server",
    "gatewayName": "maxibridge",
    "method": "tools/call",
    "paramsName": "search",
    "requestID": "2",
    "sessionID": "1f02aa20-22d8-6e87-8432-be15d4f7b5b2",
    "type": "Request"
  },
  "model": "claude-3-7-sonnet",
  "policyRefs": [
    {
      "appPolicyID": "xxx-xxx-xxx-xxx",
      "appPolicyName": "email-agent",
      "appPolicyType": "ingress",
      "matchingTeams": [
        "team-a",
        "team-b"
      ],
      "policyID": "xxx-xxx-xxx-xxx",
      "policyIdentity": "default",
      "policyName": "frontend",
      "policyNamespace": "frontend"
    }
  ],
  "principal": {
    "IP": "192.0.2.42",
    "app": {
      "component": "frontend",
      "labels": [
        "country=us",
        "another-label"
      ],
      "name": "MyApp",
      "userClaims": [
        "name=John Doe",
        "email=john.doe@acme.com",
        "@validated=false"
      ],
      "username": "john.doe@acme.com",
      "workloadGroupHash": "wg-0ff92a76a3765740e26d84947d92e5fc",
      "workloadGroupLabel": "k8s:deployment=mcp-chatbot-agent,namespace=demo",
      "workloadGroupSetHash": "wgs-0ff92a76a3765740e26d84947d92e5fc",
      "workloadGroupSetLabel": "k8s:namespace=demo"
    },
    "external": {
      "userClaims": [
        "name=John Doe",
        "email=john.doe@acme.com",
        "@validated=false"
      ],
      "username": "john.doe@acme.com",
      "workloadGroupHash": "wg-0ff92a76a3765740e26d84947d92e5fc",
      "workloadGroupLabel": "k8s:deployment=mcp-chatbot-agent,namespace=demo",
      "workloadGroupSetHash": "wgs-0ff92a76a3765740e26d84947d92e5fc",
      "workloadGroupSetLabel": "k8s:namespace=demo"
    },
    "tokenID": "1234-1224-123-1",
    "tokenName": "my-user-token",
    "user": {
      "name": "user@company.com"
    }
  },
  "provider": "openai",
  "providerType": "LLM",
  "trace": {
    "kind": "Server",
    "parentSpanID": "00f067aa0ba902b7",
    "spanEnd": "2025-03-22T14:35:00.123456789Z",
    "spanID": "6ba80aaa3b2f43d8",
    "spanName": "acuvity_prompt_input_analysis",
    "spanStart": "2025-03-22T14:35:00.123456789Z",
    "statusCode": "OK",
    "statusMessage": "Failed to make API call to service Foo.",
    "traceID": "4bf92f3577b34da6a3ce929d0e0e4736",
    "transparentSpanID": "6ba80aaa3b2f43d8"
  }
}

Changes

Changed in 4 of the 6 revisions of this API.25

  • 0f21cd12148e13See the full diff
    • ●

      added the new None enum value to the tools/additionalProperties/category response property for the response status 200

      response-property-enum-value-added

    • ○

      the category request property default value None was added

      request-property-default-value-added

    • ○

      added the new None enum value to the request property tools/additionalProperties/category

      request-property-enum-value-added

    • ○

      the category response's property default value None was added for the status 200

      response-property-default-value-added

  • e850756435a61See the full diff
    • ●

      added the new Redaction enum value to the error/stage response property for the response status 200

      response-property-enum-value-added

  • 58d2295ee11e1See the full diff
    • ○

      added the optional property latency/contentAttribution to the response with the 200 status

      response-optional-property-added

  • 010e99908a0e1See the full diff
    • ○

      endpoint added

      endpoint-added