A12 - Approval Workflow
| Group | Technology | Status | Service | Roadmap phase | Contracts |
|---|---|---|---|---|---|
| A - Governance, identity and security | PostgreSQL + API (optional Temporal adapter) | to develop | approvals | 3 | audit-event |
Role and responsibilities
Durable approval requests with separation of duties for promotions, sensitive tools and exceptions.
Data model
Tables (schema eai, every tenant table carries tenant_id uuid NOT NULL with an RLS policy on current_setting('app.tenant_id', true)::uuid): eai.approvals, eai.audit_events.
API endpoints
GET /v1/approvalsPOST /v1/approvalsGET /v1/approvals/{id}POST /v1/approvals/{id}/decide
POST /v1/approvals HTTP/1.1
Host: <control-plane-host>
X-API-Key: eai_xxx_secret
Accept-Language: en
Content-Type: application/json
{"action": "deploy", "resource_type": "agent", "resource_id": "invoice-agent", "reason": "Promote version 1.4.0 to production", "approver_roles": ["security_officer"], "quorum": 1}
{
"id": "a1ab0542-301d-4eef-a8be-64d5173ad21e",
"requester": "alice",
"action": "deploy",
"resource_type": "agent",
"resource_id": "invoice-agent",
"run_id": null,
"reason": "Promote version 1.4.0 to production",
"payload": {},
"quorum": 1,
"approver_roles": ["security_officer"],
"status": "pending",
"expires_at": "2026-10-31T01:35:52.793643Z",
"decided_at": null,
"created_at": "2026-10-30T01:35:52.792996Z",
"updated_at": "2026-10-30T01:35:52.792996Z",
"decisions": [],
"approvals_count": 0
}
Errors use the common envelope {"error": {"code", "message", "message_i18n", "details", "request_id"}}; lists are paginated with limit and cursor.
Controlled promotions
Moving a model, tool or agent to approved or active opens an approval (action: promotion, approver roles
tenant_admin and security_officer) and answers 202 with approval_id. The requester cannot approve it. On approval:
a system classified high risk (latest risk assessment high, or agent risk_level high) needs an approved risk
assessment; then the blocking promotion suites of the target run (evaluation.on_promotion); the status is applied
only when they pass. The outcome is stored in payload.promotion and audited (promotion.request,
promotion.apply, promotion.blocked). Skills (POST /v1/skills/{name}/status) and MCP servers (POST /v1/mcp/servers/{id}/status) answer 202 the same way; plugins become active through install, guarded by the owner relation; a skill can no
longer be set to approved outside an approved promotion (409 promotion_requires_approval). Retirements stay
immediate and audited. Platform wide models changed by a platform admin without a tenant are not covered, approvals
being tenant scoped.
SDK usage
Python (enterprise_ai):
from enterprise_ai import EnterpriseAI
client = EnterpriseAI(base_url="<control-plane-url>", api_key="eai_xxx_secret")
items = client.approvals.list()
TypeScript (@enterprise-ai/sdk):
import { EnterpriseAI } from "@enterprise-ai/sdk";
const eai = new EnterpriseAI({ baseUrl: "<control-plane-url>", apiKey: process.env.EAI_API_KEY });
const items = await eai.approvals.list();
Java (enterprise-ai-java):
EnterpriseAIClient client = EnterpriseAIClient.builder()
.baseUrl("<control-plane-url>")
.apiKey(System.getenv("EAI_API_KEY"))
.build();
var items = client.approvals().list();
Go (enterprise-ai-go):
client, _ := eai.NewClient(eai.WithBaseURL("<control-plane-url>"), eai.WithAPIKey(os.Getenv("EAI_API_KEY")))
items, err := client.Approvals.List(ctx)
Configuration
- Environment:
EAI_DATABASE_URL,EAI_OTEL_EXPORTER_OTLP_ENDPOINT. - Helm values:
controlPlane.enabled,controlPlane.replicas,controlPlane.resources,controlPlane.config.
Security controls
- Authentication by OIDC JWT, tenant API key with scopes, or SPIFFE mTLS; tenant resolved from the credential, never from the body.
- Authorization through OpenFGA and OPA decisions; every mutation and decision writes a hash-chained
eai.audit_eventsrow. - Secrets never returned by the API; TLS everywhere; Cilium network policies restrict who can reach the service.
Separation of duties: the requester cannot approve its own request (409 separation_of_duties).
Observability
- Metrics:
eai_approval_workflow_requests_total,eai_approval_workflow_errors_total,eai_approval_workflow_latency_seconds(Prometheus). - Traces: OpenTelemetry spans
eai.approvals.*with W3Ctraceparentpropagation, AI spans forwarded to Opik. - Audit actions:
approval.request,approval.decide,approval.expire,policy.evaluate.
FinOps aspects
Operating cost is allocated per tenant and project through labels on Usage Events and Kubernetes resource requests; no direct per-call cost.
Failure modes and resilience
- Stateless replicas behind the API Gateway, PostgreSQL HA for state; readiness probe
GET /readyzremoves unhealthy pods. - External system unavailable: fail closed for security decisions, explicit error code, no silent bypass.
Standalone adoption (client of a client)
A client of a client can adopt this brick alone with the Python SDK sub-client client.approvals (TS eai.approvals, Java client.approvals(), Go client.Approvals). Minimal deployment: Control Plane API + PostgreSQL + Keycloak (or API keys), the Control Plane loads every domain of services/control-plane/src/eai/domains; the optional chart components are switched off in deploy/helm/enterprise-ai/values.yaml (vllm.enabled, qdrant.enabled, nats.enabled, workers.enabled).
Acceptance tests
- A tenant cannot read or modify another tenant's resources (RLS test).
- Every mutation produces an audit event verifiable by
GET /v1/audit/verify. - Errors are returned in English and French according to
Accept-Language.