generated: '2026-08-02'
method: searched
source: >-
  https://docs.aigateway.cequence.ai/docs/api-registry +
  https://docs.aigateway.cequence.ai/docs/guides/multi-authentication +
  https://docs.aigateway.cequence.ai/docs/guides/security-policies +
  https://docs.aigateway.cequence.ai/docs/guides/observability
summary: >-
  Cross-cutting runtime semantics for the Cequence AI Gateway, captured from the published docs. The
  gateway is a governed MCP front door: it terminates agent authentication, applies authorization,
  rate limiting and DLP, injects upstream credentials, and audits every tool call. Idempotency is NOT
  documented — do not assume it.
surface:
  primary: Model Context Protocol (streamable HTTP)
  secondary: REST APIs registered from OpenAPI/Swagger specs and re-published as MCP tools
  no_public_rest_api: >-
    Cequence publishes no public REST API of its own for the AI Gateway control plane; the equivalent
    surface is the first-party hosted MCP server (see mcp/cequence-security-mcp.yml).
authentication:
  styles: [openid_connect_sso, oauth2_authorization_code, oauth2_client_credentials, api_key, bearer, jwt_bearer, basic, passthrough]
  broker_model: two-sided (inbound agent auth + outbound app auth); credentials are never exposed to the agent
  see: authentication/cequence-security-authentication.yml
authorization:
  model: >-
    Least-privilege endpoint/tool selection at registration time, narrowed again per Agent Persona;
    Teams map SSO groups onto MCP servers and personas.
  quote: >-
    "Only select the endpoints your agents will actually use. You can always register more later, and
    each persona can be granted an even narrower subset."
idempotency:
  documented: false
  header: null
  notes: >-
    No idempotency key, retry-safety contract or replay window is documented for the AI Gateway. Tool
    calls inherit whatever idempotency the upstream API provides. Agents should treat write tools
    (create_mcp_server, create_agent_persona, create_skill, attach_dlp_policy) as non-idempotent.
pagination:
  documented: false
  notes: >-
    No gateway-level pagination convention is published; list tools proxy the upstream/control-plane
    result sets.
rate_limiting:
  model: per-tool rolling window with method-derived defaults
  exceeded: HTTP 429 + audit reason rate_limit_exceeded
  see: rate-limits/cequence-security-rate-limits.yml
error_envelope:
  style: http-status + human-readable message + machine-readable audit reason
  problem_json: false
  see: errors/cequence-security-problem-types.yml
request_tracing:
  fields: [request.id, session_id, timestamp]
  notes: Every tool activity event carries a Request ID and Session ID.
observability:
  audit_event_fields:
  - timestamp
  - request.id
  - tool.name
  - mcp.server_name
  - user.email
  - client.ip
  - status
  - action
  - action.reason
  - duration.total_ms
  - duration.upstream_ms
  - http.status_code
  status_values: [success, error, blocked, rate_limited]
  siem_export:
  - {destination: Splunk, protocol: HTTP Event Collector (HEC)}
  - {destination: Datadog, protocol: Datadog API}
  - {destination: OTLP, protocol: OpenTelemetry Protocol (gRPC or HTTP)}
  - {destination: Syslog, protocol: TCP, UDP, or TLS}
  webhooks: >-
    No webhook or push-callback surface is documented; event delivery is SIEM export only, so no
    Webhooks/AsyncAPI artifact is emitted.
data_protection:
  dlp: Sensitive-data-protection categories and DLP policies, attachable per Agent Persona, with redaction.
  recommendation_tool: recommend_dlp_policy (deterministic)
  network_policy: Egress control per deployment pool; cell-based isolation tiers (Critical / Standard / Experimental) in private-cloud deployments.
spec_ingestion:
  accepted_formats: [OpenAPI, Swagger, YAML, JSON, WSDL, XML]
  max_size: 25 MB
  base_url: 'Operator selects the base URL — "the address the gateway will call".'
versioning:
  gateway: Continuous SaaS; no public API version segment documented.
  uap_platform: See lifecycle/cequence-security-lifecycle.yml for the UAP release train.
cross_links:
  authentication: authentication/cequence-security-authentication.yml
  errors: errors/cequence-security-problem-types.yml
  rate_limits: rate-limits/cequence-security-rate-limits.yml
  lifecycle: lifecycle/cequence-security-lifecycle.yml
  mcp: mcp/cequence-security-mcp.yml
x-evidence:
  fetched: '2026-08-02'
  urls:
  - url: https://docs.aigateway.cequence.ai/docs/api-registry
    http_status: 200
  - url: https://docs.aigateway.cequence.ai/docs/guides/multi-authentication
    http_status: 200
  - url: https://docs.aigateway.cequence.ai/docs/guides/security-policies
    http_status: 200
  - url: https://docs.aigateway.cequence.ai/docs/guides/observability
    http_status: 200