GEIST on Azure

"Why not just use Microsoft Foundry?"

Run GEIST on Foundry.

Foundry is a great PaaS — 11,000 models, 1,400 connectors, Entra Agent ID, Foundry IQ. What it doesn't ship is the application: the talent management system, the EU pay transparency engine, the multitenancy. Build that yourself for ~$1.2-2.0M and 12-18 months — or deploy GEIST on your Azure tenant in 10-14 weeks and use Foundry as the infrastructure layer.

What Foundry is great at

We run on Foundry.
Here's what we add on top.

Model catalog & routing

11,000+ models. Foundry Model Router for cost/quality tradeoffs. We use the catalog as our backbone via Azure OpenAI endpoints.

Foundry Agent Service

Stateless containerized agent runtime, Entra Agent ID per agent, built-in File Search / Code Interpreter / Deep Research. We integrate with these as tools (see roadmap below).

Foundry IQ

Managed RAG over SharePoint, Fabric, M365. We can pull Foundry IQ content into GEIST's scoped knowledge bases as a source.

Enterprise governance

Defender, Purview, Sentinel — Microsoft-native security and compliance posture. GEIST inherits all of this when deployed in your Azure tenant.

What Foundry doesn't ship

The application layer is your problem.

Microsoft is explicit about this: Foundry is a development platform. The vertical application, multitenancy, domain integrations, and compliance enforcement are the customer's responsibility. That work takes 12-18 months and ~$1.2-2.0M to build well — and you have to maintain it forever.

CapabilityFoundry aloneGEIST
Production vertical applicationsBuilding blocks only — customer assemblesTalentGEIST, PayGEIST, DermGEIST, LearnGEIST, AgentGEIST ship as deployable applications
Multitenant data isolationCustomer architects tenant boundariesPlatform → Vertical → Instance → Tenant → User scope hierarchy with row-level security, built in
Domain-specific connectors1,400+ generic connectors — wire them yourself24 connectors with domain schema knowledge baked in (PeopleXD V28, Epic FHIR R4, Workday, Salesforce, M365, SharePoint)
PII safety across providersAzure Content Safety — configure yourself, one-way redactionReversible PII tokenization — strip before LLM, restore in response. HIPAA 18-identifier patterns
Per-tenant cost trackingAggregate cost reporting onlyPer-user, per-tenant, per-app metering and quotas
Time to production12–18 months custom build (~$1.2–2.0M Year 1)10–14 weeks to deploy on customer's Azure tenant

Verified in the GEIST codebase and competitive analysis, March 2026.

How GEIST deploys on your Azure tenant

Native Azure citizen.
Your subscription. Your data. Your EA.

GEIST's architecture already abstracts cloud dependencies. Most of the migration is configuration; only secret management and blob storage need new adapters (~3-4 weeks engineering). The full deployment plan is below.

ComponentAzure equivalentNotes
Container orchestrationAzure Container Apps (ACA) or AKSGEIST uses SSE — ACA viable; AKS for max control
DatabaseAzure PostgreSQL Flexible Server + pgvectorWire-compatible, zero schema migration
Cache / queueAzure Cache for Redis (Premium P1)Required for Streaq Streams
Object storageAzure Blob StorageAzureBlobBackend adapter (~1 week)
SecretsAzure Key VaultAzureKeyVaultClient adapter (~2-3 weeks)
AI inferenceAzure OpenAI / Foundry catalogAzureOpenAIEmbedder ships today (3072 dims, exact match)
SSO / identityMicrosoft Entra IDSAML 2.0 SP fully implemented — config only
MonitoringAzure Monitor + Application InsightsOTLP endpoint config only

10–14 wk

Migration time (2 engineers)

Low cost

Azure infra cost (ACA, EU region)

EU West

Primary region — GDPR compliant

What we ship vs what's on the roadmap

Where we're honest about gaps.

Production today

  • Azure OpenAI as a routed provider (3072-dim embeddings, exact match — no re-indexing)
  • Read-only calls to pre-published Foundry agents
  • Microsoft Entra ID via SAML 2.0 (per-tenant IdP config)
  • GEIST RAG with 5-level scope enforcement
  • Reversible PII tokenization across all providers

Alpha

  • Foundry Tools — File Search, Code Interpreter, Deep Research
  • Foundry IQ as a knowledge source adapter
  • Entra Agent ID per GEIST agent for centralized governance
  • Hallucination/groundedness detection via Content Safety Groundedness API
Where we're honest about gaps: We don't claim Foundry Tools or Entra Agent ID integration today — they're scoped engineering work, not shipped capability.

See the deployment analysis for your Azure tenant.

We map every GEIST component to your Azure SKU set, estimate the migration timeline, and identify which Foundry services to adopt vs keep in GEIST. 30-minute architecture review — bring your IT and procurement leads.