Nine planes.
One coherent platform.
Modern data work used to mean stitching together a pipeline tool, a warehouse, a catalog, a governance layer, a BI tool, an observability stack, and an AI library. Each one with its own login, its own RBAC, its own audit trail, its own bill. D-Pulse is the opposite design: nine purpose-built planes inside one platform, sharing one identity model, one governance graph, and one operational surface.
Each plane is independent enough to scale on its own, and integrated enough that the seams disappear. Below is what each plane does, and why it's a plane instead of a feature bolted onto something else.
01Pipeline Plane
How data movesThe visual workflow engine. Drag-and-drop builder with 56+ node types, n8n-style expression editor, and a backend-driven node registry that makes new connectors live without redeploys. Templates cover medallion ETL, CDC replication, data quality gates, and reverse ETL out of the box. Scheduling is cron-based with dependency awareness, backfill, retries, and SLA tracking.
02Lakehouse Plane
Where data livesBronze → Silver → Gold medallion architecture on object storage. Supports Iceberg, Delta, and Hudi table formats — all queryable via DuckDB and Trino. Bronze holds raw landing, Silver holds cleaned and typed data, Gold holds business-ready marts. The plane manages partitioning, compaction, schema evolution, and retention. You can run on its native lakehouse or point it at Snowflake / BigQuery / Databricks.
03Catalog Plane
What data existsUnified data catalog with column-level lineage, automated profiling, and search. Crawlers populate metadata from every connected source — warehouses, lakehouses, BI tools — into a single graph. Lineage flows through pipelines, transforms, and downstream products. Profiling runs row counts, null rates, distinct values, and pattern detection on a schedule. The catalog is the source of truth for what data exists, where it came from, and what depends on it.
04AI Plane
How data is queried, generated, and reasoned aboutMulti-provider AI with 9 backends — Claude, OpenAI, Gemini, Azure OpenAI, Ollama, DeepSeek, Mistral, Groq, Custom. Powers NL→SQL, pipeline generation, root cause analysis, page-context-aware chat, and tool-using agents (MCP). Embedded AI features (ghost nodes, auto-fill, smart schedule, error diagnosis) live inside other planes. Ollama-first means you can run the entire AI plane locally with no data leaving your VPC.
05Governance Plane
Who can do whatRBAC, data contracts, classification, and policy enforcement. 5-tier role hierarchy (Super Admin → Workspace Admin → Data Engineer → Analyst → Viewer) with workspace-level filtering. Field-level data classification (L1-L4) drives masking and access policies automatically. Policies are written declaratively and enforced via OPA at query time. Compliance reports for SOC 2, HIPAA, GDPR are generated from the audit trail.
06Observability Plane
Is it healthyMetrics, logs, traces, and SLAs across every plane. Built on Prometheus + Grafana + Loki + Marquez under the hood, exposed as a single Operations page. SLA engine tracks freshness, completeness, and accuracy contracts per dataset. Platform Health Score aggregates plane-level health into one number leadership can read. Alerting routes to Slack, Teams, PagerDuty, and email with deduplication.
07Serving Plane
How data leavesData products, REST API publishing, QueryLab, and data marts. Authors define a dataset as a product, set the contract, set the consumers, and the plane handles versioning, subscriptions, and metrics. Published datasets become governed REST APIs with rate limiting and auth. QueryLab is the SQL workbench for analysts — DuckDB-backed with Redis caching. Data marts are materialized gold-layer views with refresh policies.
08Connector Plane
What data plugs inFirst-party connectors across 12 categories: databases, warehouses, object storage, streaming, CDC, vector DBs, SaaS APIs, files, time-series, and more. Built on a REST connector framework that wraps any HTTP API as a first-class node — most teams build a new connector in an afternoon. CDC handled via Debezium for Postgres, MySQL, SQL Server, MongoDB, and Oracle. Schema registry tracks contract changes and prevents silent breaks downstream.
09Execution Plane
How work runsWorkload management, priority queues, engine routing, and quotas. Different workload classes (interactive, batch, ML training) get different resource pools and priorities. The execution plane decides whether a query runs on DuckDB, Trino, or a federated engine based on size, latency requirements, and cost. Backpressure protects downstream systems from runaway loads. Concurrent session control and seat management round out the operational surface.
Why planes, not features
The temptation when building a data platform is to ship a pipeline tool first, then bolt on a catalog, then bolt on AI, then bolt on governance. The result is a product that works at the demo level and falls apart in production: features that don't share identity, don't share lineage, don't share policy, and can't be operated as one system.
The 9-plane model is the opposite. Each plane is its own service with its own data model, its own API, and its own scaling profile. The Pipeline plane can scale independently from the Lakehouse plane. The AI plane can be swapped out (Claude → Ollama → custom) without touching anything else. The Governance plane sees every action across every plane, because it's wired into the request path.
The cost of this design is up-front complexity — nine planes is more to build than one monolith. The benefit is that the system stops being fragile once you put real data and real users on it. When a plane needs to change, only that plane changes. When a plane needs to scale, only that plane scales. When a plane fails, the others degrade gracefully instead of cascading.
That's the bet behind D-Pulse: the right number of moving parts is more than one and less than fifteen. Nine is what it took to cover the work without leaving seams.
See it in action
The architecture only matters if it solves a real problem for your team. The fastest way to find out whether D-Pulse fits is a 30-minute walkthrough with your actual stack on screen.