Deployment · Architecture · Operations
Choose where to run your customer data platform (CDP): your datacenter, your cloud account or both. Assign an owner for hosting, access, backups and updates. Test the setup before using real customer data.
Single-tenant · Customer cloud / on-premise / sovereign
PrivyCDP Core
Identity Resolution
AI Decision Layer
Event Bus
Vector Store
Audit Ledger
Section 01 · Deployment Models
Each setup serves one customer in their own systems. Scope hosting, hardware and operating work separately from the software licence.
Run the software in your own datacenter. Your team or a chosen partner operates it. Ask about the requirements for an isolated network.
Run the software in your AWS, Azure, GCP or OCI account. Check the network paths, access rules and any outbound requests.
Keep core records in your datacenter and selected services in your cloud account. Test data movement and timing across regions.
Assess a sovereign cloud or an isolated network. Agree how updates reach the system and test the rules for your workload.
Section 02 · Architecture Overview
Plan three parts of the setup. Assign capacity, access and audit checks to each one. Your team owns the systems that run them.
Edge
Ingestion · activation · external integration interface
EDGE-01
Event Collector
SDK / HTTP / Kafka
EDGE-02
Activation Gateway
Reverse-ETL / Webhook
EDGE-03
Integration Bus
MCP-ready connectors
Core
Identity · decisioning · AI inference · orchestration
CORE-01
Identity Resolution
Deterministic + probabilistic
CORE-02
Decision Engine
Real-time rules + ML scoring
CORE-03
AI Decision Layer
Private LLM · MCP
CORE-04
Journey Orchestrator
Workflow runtime
CORE-05
Segmentation Service
Streaming + batch
CORE-06
Policy Engine
Consent · residency · RBAC
Storage
Persistence · audit chain · vector indices
STORE-01
Profile Graph
Customer-owned DB
STORE-02
Event Log
Immutable, append-only
STORE-03
Vector Store
Embeddings · features
STORE-04
Audit Ledger
Hash-chained
Section 03 · DevOps & Update Model
Updates are delivered as versioned, signed artifacts. The customer's DevOps team controls the promotion of every release through a deterministic five-phase pipeline. There is no forced auto-update and no shared SaaS rollout.
Stage
Versioned release artifact is pulled into the customer's staging environment. Signed, hash-verified, and reproducible.
Canary
Release is exposed to a controlled traffic slice inside the customer boundary. Telemetry is observed locally — no telemetry egress to PrivyCDP.
Audit
Customer security and compliance teams review change manifests, dependency diff, and policy impact reports before promotion.
Promote
Promotion to production is executed by the customer's DevOps team on the customer's schedule. Roll-forward and roll-back are deterministic.
Verify
Post-deployment verification runs against the customer-owned audit ledger. Version, signature, and policy state are recorded immutably.
Section 04 · Support Boundaries
PrivyCDP, the customer, and joint engagements own distinct interfaces. There is no overlap and no implicit responsibility — every boundary is documented in the deployment runbook.
Section 05 · Infrastructure Separation
Infrastructure, DevOps, cloud, and hardware are scoped separately from the software license. This separation is intentional — it keeps total cost of ownership transparent, prevents vendor lock-in to a specific cloud, and ensures the customer retains full operational control of the environment in which PrivyCDP runs. Architecture review, deployment partnership, and managed support are available as distinct enterprise engagements.
Planning your evaluation
Use workload assumptions to size the environment. This is a planning checklist; a fixed hardware minimum or production SLA needs a configuration-specific proposal.
Provide average and peak events per second, payload size, retention, profile count, concurrent queries, activation volume and latency objectives. Reserve capacity for replay and recovery; benchmark synthetic peak load before acceptance.
For a VPC, identify cloud-account ownership, regional services, networking and managed-service permissions. On-premise, include hardware capacity, storage growth and replacement responsibility. Hybrid deployments also need measured inter-site latency and explicit transfer rules.
Approve the runtime and database versions for the selected release. Document DNS, TLS, secrets, identity-provider integration, inbound routes and allowed outbound destinations. A deployment diagram alone does not validate those prerequisites.
Load synthetic data into staging, test one integration end to end and restore from backup. Agree recovery point and recovery time targets with named owners. Rehearse rollback and migration recovery before approving the first production release.
Related engineering context: Privacy-first data-flow planning. Use it with the page-specific evaluation questions above.
Discuss your deployment with engineers.
Bring your data flow, operating owners and evaluation questions.
Published: · Updated:
Editorial dates describe content changes. Confirm current sources before making legal, security or purchasing decisions.