Architecture

Control plane and data plane, cleanly separated.

The control plane orchestrates on metadata only; the data plane runs inside your environment, where your data stays. Run data planes across your clouds, VPCs and on-prem — all dialling out to one control plane.

FabriCloud Control Planemanaged or self-hostedorchestration · governance · observability — metadata & control signals onlyPolicyIdentityGatewayObservabilityDeploycontrol signalsmetadata only · no raw dataYOUR ENVIRONMENTNO RAW DATA LEAVESFabriCloud Data Planeruns in your env · holds data & indexesComputeConnectGraphNetworkDatabasesWarehousesLakesStreamsSaaS appsread / write in placefabricloud.ai
FabriCloud platform — control plane (metadata only) and data plane (in your environment)
FabriCloud Control Planemanaged or self-hostedYOUR AWSFabriCloudData PlaneConnectComputeGraphNetworkYOUR VPCFabriCloudData PlaneConnectComputeGraphNetworkON-PREMFabriCloudData PlaneConnectComputeGraphNetworkoutbound-only · mTLS ↑Data planes dial out. No inbound ports on your data — the control plane never connects in. Air-gapped supported.fabricloud.ai
Deployment topology — outbound-only, mTLS, no inbound ports on your data
Control + data plane

Two planes, cleanly separated.

Control decisions flow down. Raw data never flows up.

In depth

The two planes, in depth.

Control plane

FabriCloud Control Plane

gRPC control channel · mTLS · desired-state reconciliation

The management layer. It compiles policy, plans and reconciles deployments, brokers identity, exposes every API and aggregates telemetry — operating only on metadata, configuration and control signals. Run it as a managed service or self-host it in your own tenant; in neither case does it store or process your data. A continuous reconciliation loop drives each data plane toward its declared desired state.

  • Managed or self-hosted
  • Metadata & config only — never data
  • Desired-state reconciliation
  • Multi-data-plane fleet management
  • Stateless, horizontally scalable
  • Signed, auditable control channel
Data plane

FabriCloud Data Plane

Kubernetes-native · regional sharding · offline-tolerant

The runtime that lives inside your environment. It hosts the connectors, the query and compute engines, the catalog indexes and the local policy-enforcement points, executing compiled policy and work specifications against data in place. One control plane can govern many data planes — per region, per business unit, per sensitivity tier — each fully isolated.

  • Runs in your cloud / VPC / on-prem
  • Single-tenant per deployment
  • Local policy-enforcement points
  • Holds data & indexes in your env
  • Many data planes per control plane
  • Operates through control-plane outages
Control plane

The brain — five services, zero data.

Everything that decides, governs and exposes. None of it touches a byte of your data.

5 services
Control plane

FabriCloud Policy

declarative · OPA / Rego-compatible · query-time

Governance is declarative and centralized. You express access, masking, residency, retention, classification and quality rules as version-controlled policy; the control plane compiles them and distributes enforcement to every data plane, where they are evaluated at query time, at the source. Policies are testable, auditable and promoted through environments like application code.

  • Policy-as-code, stored in git
  • Access · masking · residency · retention
  • Query-time enforcement at source
  • Classification- & tag-driven rules
  • Dry-run & impact simulation
  • Environment promotion & rollback
Control plane

FabriCloud Identity

SAML / OIDC · SCIM 2.0 · OPA-compatible

FabriCloud federates to your identity provider rather than becoming a parallel one. Users and groups arrive through SSO and SCIM; entitlements are expressed as role- and attribute-based rules evaluated at query time, down to row, column and cell. Service and agent identities are first-class, issued short-lived, narrowly scoped credentials with full session logging.

  • SAML / OIDC single sign-on
  • SCIM user & group provisioning
  • RBAC + ABAC
  • Row, column & cell security
  • Dynamic data masking
  • Short-lived service & agent identities
Control plane

FabriCloud Gateway

MCP · GraphQL · OpenAPI

FabriCloud is MCP-first: every governed dataset, connector and tool is exposed to AI agents as native Model Context Protocol tools, so agents discover and invoke them with access enforced at the source. The same surface is available as a uniform GraphQL and REST API with typed SDKs — the web console is just another client.

  • MCP-first — native tools for agents
  • GraphQL + REST
  • Typed SDKs (Python, TS, Go, Java)
  • Webhooks & event streams
  • Versioned, backward-compatible
  • Scoped, revocable tokens
Control plane

FabriCloud Observability

OpenTelemetry · Prometheus · OpenLineage

Every layer emits metrics, traces, logs and data-quality signals, tied together by lineage from source column to consumer. Operational telemetry exports to your monitoring stack; access and policy events stream to your SIEM as an immutable audit trail. Built-in dashboards cover freshness, cost, query performance and every policy decision made.

  • Metrics / traces / logs (OTel)
  • Column-level lineage
  • Immutable, exportable audit log
  • SIEM streaming
  • Data-quality & freshness SLOs
  • Cost & usage attribution
Control plane

FabriCloud Deploy

Terraform · Helm · marketplace

Standing up FabriCloud is a templated, repeatable operation. Terraform modules and Helm charts provision a hardened data plane into your account with compliant defaults; cloud-marketplace listings handle billing and entitlement. Everything is GitOps-friendly, so environments are reproducible, reviewable and upgradeable — no snowflake servers.

  • Terraform modules
  • Helm charts
  • AWS / Azure / GCP marketplaces
  • On-prem & air-gapped installer
  • Hardened, compliant defaults
  • GitOps & automated upgrades
Data plane

The muscle — four services, in your environment.

Everything that holds and processes data. It runs where your data already is.

4 services
Data plane

FabriCloud Compute

Kubernetes / Ray · SQL pushdown · GPU scheduling

A scheduler that places query, transformation, ML-feature and agent workloads onto compute in your own account, scaling pools up and down with demand. Workloads are isolated from one another, pushed down to where the data sits to avoid movement, and can target heterogeneous compute — CPU pools, GPU pools, serverless, or your existing warehouse engines.

  • Elastic autoscaling
  • Per-workload isolation (namespaces)
  • Pushdown to source engines
  • CPU / GPU / serverless targets
  • Spot- & reserved-capacity aware
  • Queue priorities & quotas
Data plane

FabriCloud Connect

CDC · schema-on-read · open table formats

Native, bi-directional connectors that read and write data in place — no raw data is copied into a proprietary store. Each connector understands the source’s schema, partitioning, change feed and access model, so FabriCloud can index, govern and query without lift-and-shift. Connectors run inside the data plane and authenticate with your own credentials.

  • Read & write in place
  • Change data capture (CDC)
  • Schema discovery & drift handling
  • Open table formats (Iceberg / Delta / Hudi)
  • Pushdown predicates & partitions
  • Credentials stay in your env
Relational
PostgreSQLMySQLSQL ServerOracleMariaDB
Warehouses
SnowflakeBigQueryRedshiftSynapse
Lakehouse & tables
DatabricksApache IcebergDelta LakeApache Hudi
Object storage
Amazon S3Google Cloud StorageAzure BlobMinIO
Streaming
Apache KafkaAmazon KinesisGoogle Pub/SubConfluent
SaaS & apps
SalesforceWorkdayServiceNowSAPMicrosoft 365
Data plane

FabriCloud Graph

entity resolution · knowledge graph · catalog

The engine behind the Knowledge Fabric. It resolves entities across sources, builds and stores the relationship graph, and maintains the searchable catalog and lineage — all as indexes held inside the data plane. It turns connected data into context that analytics, search and agents can query directly, without moving anything out.

  • Entity resolution & matching
  • Relationship / knowledge graph
  • Searchable catalog & metadata
  • Column-level lineage
  • Vector & semantic indexes
  • Indexes stay in your env
Data plane

FabriCloud Network

mTLS · outbound-only · PrivateLink

The data plane runs entirely within your network perimeter with no public ingress to data. Control traffic and data traffic are separated, and the only outbound path is a signed, outbound-initiated control channel — there are no inbound ports on your data. Private connectivity options keep everything off the public internet.

  • No public ingress to data
  • Outbound-only control channel
  • VPC peering / PrivateLink / PSC
  • Egress allow-lists
  • Private DNS resolution
  • Air-gapped mode
Reliability

Private-cloud reliability, enterprise support.

FabriCloud is operated like infrastructure, not a side project. The control plane is monitored around the clock and data planes keep serving through control-plane interruptions. Automated backups, disaster-recovery targets, a defined incident-response process, tiered support and scheduled upgrade windows mean the platform behaves predictably in production — and your team always knows what to expect.

Availability

99.9% uptime target

Control-plane SLA; data planes keep serving through control-plane interruptions.

Monitoring

24ƥ7 monitoring

Continuous health checks and a public status page for every region.

Backups

Automated backups

Encrypted, scheduled backups of metadata and indexes you can restore.

Disaster recovery

DR with RPO / RTO targets

Multi-AZ by default, with documented recovery-point and recovery-time targets.

Incidents

Severity-based response

A defined incident-response process with status updates and post-incident reviews.

Support

Standard · Business · Enterprise

Tiered support with response-time SLAs and named contacts at the top tier.

Upgrades

Scheduled upgrade windows

Versioned, backward-compatible releases applied in windows you control.

Patching

Security-patch SLAs

High/critical findings remediated to agreed timelines, validated before rollout.

Deployment topology

One control plane, many environments.

Run data planes across your clouds, VPCs and data centres — all dialling out to a single control plane.

Bring the cloud to your data.

See it deployed in your own environment — analytics, apps and agents on governed data — a boundary raw data never crosses.

Start building