VarolioVarolio Docs
Deployment scenarios

Cloud SaaS

Multi-tenant deployment operated entirely by Varolio, hosted in the EU

Cloud SaaS

Cloud SaaS is Varolio's multi-tenant deployment model. A single environment, operated entirely by Varolio, serves multiple customers; tenants are isolated logically at the application layer and share the same managed infrastructure in Varolio's AWS account.

Cloud SaaS removes operational burden: no cloud account to provision, no networking to configure, no upgrade cycle to schedule. Customers get the same release cadence and operational telemetry that Varolio uses internally.

Customers
Workspace A
Workspace B
Varolio AWS Account · eu-central-1 (Frankfurt)
Proxy
AWS WAF
AWS WAF
Application Load Balancer
Application Load Balancer
Istio Ingress Gateway
Istio Ingress Gateway
Object storage
Amazon S3
Amazon S3
keys namespaced per workspace
Amazon EKS cluster (shared)
Engine
API service · control-plane access
Agent
AI orchestration · retrieval
Worker
Temporal workflow worker
MCP Server
MCP endpoint
External Managed Services
Supabase
Supabase
PostgreSQL
Elastic Cloud
Elastic Cloud
Redis
Redis
Temporal Cloud
Temporal Cloud
workflow orchestration
Vertex AI (Gemini)
Vertex AI (Gemini)
Amazon Bedrock
Amazon Bedrock

Region and environment

Cloud SaaS is hosted on AWS in eu-central-1 (Frankfurt) — the platform is EU-resident by default. The production environment has its own VPC, EKS cluster, and secret store, with externally managed services: Supabase (PostgreSQL), Elastic Cloud, Redis, and Temporal Cloud (workflow orchestration). LLM inference runs against Vertex AI's europe-west3 (Frankfurt) endpoint as the primary provider.

Tenancy and isolation

Tenant isolation is enforced at the application layer on every request:

  • Database and search queries are filtered by workspaceId. A query that does not carry a workspace scope is rejected at the API boundary.
  • Object-storage keys are namespaced per workspace. Cross-tenant key access is not possible because the API never issues a key for a workspace other than the caller's.
  • Integration credentials (Microsoft 365, Gmail, WhatsApp, CRM connections) are stored per tenant, encrypted in AWS Secrets Manager, and resolved at runtime for the requesting workspace only.
  • Access control is JWT-based with account-, workspace-, and inbox-scoped RBAC. The admin surface runs separately from the user-facing application.

Public surface

HostnamePurpose
app.varolio.ioMain user web application (Studio)
api.varolio.ioPlatform API (engine)
mcp.varolio.ioMCP server for AI-tool access

The API edge is fronted by AWS WAF (AWS managed rule groups — IP reputation, common rule set, known bad inputs, SQLi — plus per-IP rate limiting) in front of an internet-facing Application Load Balancer that enforces client-side TLS with an ACM certificate. Traffic then flows to the Istio ingress gateway inside the EKS cluster.

Encryption

LayerCloud SaaS
Supabase (PostgreSQL) at restAES-256, provider-managed
Elasticsearch at restAES-256, provider-managed
S3 bucketsSSE-S3 (AES-256)
Application-layer secrets and integration credentialsAWS Secrets Manager, synced to the cluster via the Secrets Store CSI driver
In transitTLS 1.2+ at every external hop

Operational ownership

AreaOwner
AWS account, VPC, EKS, data storesVarolio
OS and node patchingVarolio (EKS-managed node groups)
Application code and releasesVarolio (GitOps-driven rolling deploys)
Database backupsVarolio
Secret rotationVarolio
Monitoring, alerting, on-callVarolio
Customer user accounts and rolesCustomer
Customer SSO configurationShared — customer provides IdP (OIDC), Varolio configures
Integration credential authorizationCustomer (authorizes OAuth scopes / provides credentials)

AI and embedding providers

LLM calls and embedding generation use the default AI providers documented on the Application Architecture page: Google Vertex AI (Gemini, primary), with Anthropic Claude via Amazon Bedrock as secondary. Customers who need inference to run against their own model deployment should look at BYO Data Store.

Upgrade cadence

Every change to production goes through the same release pipeline: a passing CI suite, peer-reviewed pull-request approval, and GitOps-driven deployment. Deploys are rolling updates — the API tier runs with multiple replicas so customer requests are not interrupted during a deploy, and rollback to the previous revision is available if a regression surfaces.

When to choose a different model

  • Choose Dedicated Cloud if you need single-tenant isolation and a choice of region, with Varolio still operating everything.
  • Choose BYO Data Store if your content data and search index must live on infrastructure you control.
  • Choose Cloud-Prem if the entire platform must run inside a cloud account you own.

AI Tools

Ask ChatGPTAsk Claude

On this page