Platform Architecture

This page describes how Nexa is organized conceptually — the layers data passes through and how they relate — without going into deployment internals. Read it to build a mental model before you start clicking through screens. For install topology and infrastructure, see the Operate section instead.

The flow at a glance

Data enters through ingestion, moves up the medallion layers via generated pipelines, and is consumed by agents, automations, and apps. Governance and cost tracking observe every layer.

        Sources (DBs, queues, files)
                  │  ingestion (connections + connectors)
                  ▼
        ┌───────────────────────────┐
        │  RAW  (bronze)            │  source data landed as-is
        └───────────────────────────┘
                  │  generated pipeline
                  ▼
        ┌───────────────────────────┐
        │  CURATED  (silver)        │  cleaned, conformed entities
        └───────────────────────────┘
                  │  generated pipeline
                  ▼
        ┌───────────────────────────┐
        │  CONSUMPTION  (gold)      │  wide, analysis-ready entities
        └───────────────────────────┘
                  │
                  ▼
        Agents · Automations · Apps · BI

        ── Governance & cost tracking span all layers ──

Ingestion

Ingestion is how source data gets into Nexa. It has two concepts:

• Connections hold the endpoint and credentials for one source system (relational database, message queue, or file storage). They are defined once and reused.
• Connectors bind a connection to specific tables or files and a sync schedule, in either live (continuous) or batch (scheduled) mode.

The output of ingestion is data landed in the Raw layer. See Connections and Connectors.

The medallion layers

Nexa models data in three layers. The value of the model is that each boundary is an explicit, governed mapping — not ad-hoc transformation code.

Layer	Also called	Contents	Who defines it
Raw	Bronze	Source data landed unchanged	Ingestion (automatic)
Curated	Silver	Cleaned, deduplicated, typed business entities	Curated-layer mappings
Consumption	Gold	Joined, wide, analysis-ready entities and metrics	Consumption / Canvas mappings

You define the Raw → Curated mapping in Data Flow and the Curated → Consumption mapping in Canvas. See Curated layer and Consumption layer.

Tip

The mappings are the source of truth. Nexa treats “which field becomes which” as a declared definition and generates the pipeline from it, rather than you writing transformation code by hand.

Generated pipelines

Between the layers, Nexa compiles your saved mappings into runnable pipelines and schedulable jobs. This is the “magic” step: you declare intent (the mapping), and Nexa produces the executable code and runs it on your data platform’s compute.

• Pipelines materialize a layer from the one below it.
• Jobs schedule and run pipelines, with run history you can inspect.
• Projections produce derived, materialized shapes on top of consumption data.

See Pipelines overview, Jobs, and Projections.

Agents and automations

On top of consumption-ready data, Nexa runs two kinds of AI workload:

• Agents — AI that answers questions or performs tasks against your governed data. Agents have a lifecycle (draft → validated → deployed) and run in environments such as dev, staging, and prod. See Agents.
• Agentic automations — Multi-step workflows built from triggers and actions. A plan is generated and validated before it runs, and workflows are promoted through environments. See Automations.

Deployed agents are surfaced to end users through apps — for example a Streamlit or Slack front end — managed on the Apps screen.

Governance and cost

Governance is a cross-cutting layer, not a stage in the flow. It observes and constrains everything above:

• Business glossary — shared definitions of business terms and entities. See Business glossary.
• PII tagging — sensitive fields identified and tagged during mapping.
• Change review — structural changes routed through approval. See Review changes.
• Cost tracking — usage and spend across layers. See Cost tracking.

Where it runs

Nexa is customer-managed. The web UI and backend services run in your cloud (AWS or Azure), and the pipelines and compute they drive run in your data platform (Snowflake or Databricks). Everything Nexa creates lives in your own account. For the deployment and infrastructure view, see Reference architecture and Deployment model.

Next

QuickstartWalk the whole flow hands-on, source to pipeline.

Data integrationModel the Raw, Curated, and Consumption layers.

PipelinesHow mappings become runnable pipelines and jobs.

AgentsBuild AI on top of governed data.