SYLVA Labs · AI Pulse

Azure services, data & security

A customer technical overview of what AI Pulse deploys into your Azure tenant, the data it reads, the permissions it needs, and how your information stays inside your own environment.

Product AI Pulse — AI Assurance Platform Delivery Azure Marketplace Managed Application Audience Customer IT / Security / Cloud teams Status Pre-release draft

Contents

On this page

1How AI Pulse is deployed

AI Pulse is delivered as an Azure Marketplace Managed Application. When you install it, it deploys entirely into your own Azure subscription — a single, self-contained set of resources in one resource group. There is no shared multi-tenant service: each customer runs their own isolated instance.

AI Pulse then reads signals from the AI and security services already in your environment (read-only) to build its governance dashboards. It does not run your AI models, and it does not move your data out of your tenant. Isolation is by deployment boundary — your instance, your subscription, your identity.

2Prerequisites

Have these ready before installing. Items marked optional only affect specific dashboards, which show an honest "not connected" state until enabled.

3What gets created in your tenant

A single resource group containing these resources — all standard, all in your subscription:

ResourcePurpose
Container App — portalThe AI Pulse web application itself.
Container App — LangfuseSelf-hosted AI-usage tracing. Keeps agent telemetry inside your tenant.
Container Apps environmentThe managed host both apps run in.
PostgreSQL flexible server (2 databases)Stores the audit log, governance records, and Langfuse data.
Key VaultHolds the deployment's secrets (database password, gateway key, tracing key).
Storage account + file sharePersistent working storage for the portal.
Managed identityThe read-only identity the portal uses to gather signals (see §5).

4Azure & Microsoft services it reads

Every connection below is read-only. This is the telemetry AI Pulse gathers and what each signal powers in the product.

included works with the standard install optional needs a service or licence you enable
ServiceWhat it readsWhat it powers
Azure Resource GraphInventory of AI resources — Azure OpenAI, AI Foundry / ML, Cognitive Services, Search, Bot Service, model deployments.The AI asset registry — a live map of every AI system you run.included
Azure Cost ManagementMonth-to-date spend by service, daily AI-model spend, cost anomalies.Cost & ROI dashboards (FinOps for AI).included
Azure PolicyCompliant / non-compliant resource counts and the named resources.Compliance posture.included
Microsoft Defender for CloudSecure Score, security recommendations, active alerts, regulatory-standard results.Security posture and attack-surface views.optional
Azure Monitor — Activity LogRecent administrative changes to AI resources (who changed what).The admin change trail.included
Azure Monitor — MetricsPer-service call volume, latency, token counts.Usage and performance widgets.included
Log Analytics / Microsoft SentinelDeep per-agent traces and security-analytics signals.Advanced telemetry and threat views.optional
Azure AI Foundry / Azure MLModel evaluation results (groundedness, relevance, safety, etc.).Model quality and drift tracking.included
Azure AI Content SafetyPrompt-shield and harm-analysis checks against your agents' prompts.Content-safety posture.optional
Azure Advisor + Logic AppsOptimization recommendations; automation-workflow run history.Operational-health widgets.included
Microsoft Entra ID (Graph)Directory names for access assignment, licence/SKU detection, and unsanctioned ("shadow") AI app discovery.Identity, licensing, and shadow-AI discovery.included
Microsoft 365 — Copilot auditMicrosoft 365 Copilot usage events.Copilot adoption & usage.optional
Microsoft PurviewSensitivity labels and data-security signals.Data-governance coverage.optional
Power PlatformCopilot Studio agents and Power Automate flows.Low-code / citizen-developer AI discovery.optional
Self-hosted Langfuse (in your tenant)Per-agent call / token / cost / latency, plus an in-tenant scan for exposed personal data (counts only — never stores the data itself).Agent-level observability.included

5Permissions it needs (and why)

AI Pulse follows least-privilege: it only ever asks for read access, and never gets Contributor, Owner, or user-admin rights. Permissions fall into three groups.

Granted automatically during install (Azure roles)

RoleScopeWhy
ReaderSubscriptionSee resources and their configuration to build the AI inventory.
Cost Management ReaderSubscriptionRead spend figures for the cost dashboards.
Key Vault Secrets UserThe app's own Key VaultRead its own deployment secrets at start-up.

Granted in one post-install step (Microsoft Graph, read-only)

Granted to the app's managed identity after deployment (the identity doesn't exist until then). All read-only:

PermissionWhy
Organization.Read.AllDetect licences (Copilot, Purview entitlement).
User.Read.All · Group.Read.All · Directory.Read.AllLet admins assign access by typing a person's email or a group name (not just IDs).
AuditLog.Read.AllAdmin audit trail and Copilot activity.
Application.Read.All · DelegatedPermissionGrant.Read.AllDiscover unsanctioned ("shadow") AI apps consented in your tenant.

Optional — you grant only if you want the matching widget

How to read a dark widget. If an optional service isn't connected, that dashboard shows a clear "not connected / permission pending" state rather than a blank or a fabricated number. Nothing is ever invented to fill a gap.

6Data residency & isolation

Your data stays in your tenant. Here is exactly how — stated plainly, including the one nuance.

Full transparency — the one outbound call. To confirm your subscription is active, the portal makes a single lightweight check to SYLVA's licensing service. It sends only your tenant ID and your subscription key — no customer data, no telemetry, no usage records. It is cached and tolerates outages. This is the sole SYLVA-facing connection after install.
Current network posture (be aware). In this release, the app's database and Key Vault are reachable over the Azure backbone with public network access enabled (protected by credentials and firewall rules). Locking these behind private endpoints and network isolation is an actively planned hardening step. If your policy requires private-endpoint isolation on day one, raise it with SYLVA before deployment. roadmap

7Installation at a glance

A typical install is five steps. SYLVA can run these with you.

  1. Register sign-inRun the connector script once — it creates the Entra sign-in app and records who the first administrator will be.
  2. Deploy from MarketplaceLaunch AI Pulse from Azure Marketplace and complete the short wizard (region, a database password, your gateway and subscription keys).
  3. Point sign-in at the appRun the redirect-URI script with the address the deployment produced, so single sign-on works.
  4. Grant the read-accessRun the Graph-access script once so the identity, licence, audit, and shadow-AI widgets light up.
  5. First sign-inThe first administrator signs in and completes the in-product setup wizard.

8What works out of the box

Included immediately

AI asset inventory, cost & ROI, Azure policy compliance, admin change trail, usage & performance, model-quality, identity & shadow-AI discovery, and in-tenant agent observability.

Needs a service you enable

Defender / Sentinel security depth, Content Safety coverage, Microsoft 365 Copilot usage, and Purview data-governance signals — each lights up when you connect the matching service or licence.

9The full documentation set

This overview is the anchor document. A complete customer pack usually adds the pieces below — we can produce each on request.

DocumentPurpose
Azure services, data & security (this document)The technical anchor — what deploys, what's read, what's needed.ready
Installation & onboarding guideStep-by-step install with screenshots and the exact scripts.ready
Security & trust white paperThreat model, data-flow diagram, and control mapping for security reviews.ready
Data processing & privacy statementWhat data is processed, where, and for how long — for DPO / legal review.ready
Permissions referenceEvery role and Graph permission with its exact identifier, for change-control.ready
Operations & support guideBackup, updates, monitoring, and how to get help.ready

SYLVA Labs Pte Ltd · AI Pulse — AI Assurance Platform. Customer technical overview, pre-release draft.

Prepared for review. Figures and service names reflect the current shipping install package; items marked roadmap are planned, not yet shipped, and are labelled as such deliberately.