Overview

Measured is the data quality foundation for AI. It continuously validates the data going into your models, training sets, feature tables, RAG sources, and the data coming out of them, so the systems you build on AI run on data you can trust. It's built on the open-source GX Core engine, with a native MCP server so your own LLMs and agents can participate directly.

Why data quality is an AI problem

AI fails quietly. A null where there shouldn't be one, a silent schema drift, a duplicate that skews a feature, none of it throws an error. It just degrades every model, agent, and answer downstream. Measured puts a continuous, measurable quality gate around the whole AI loop:

  • Quality in, validate the inputs your models depend on before they reach the model.
  • Quality out, treat AI-generated records and scores as first-class data assets and validate them too.
  • Your LLMs, plugged in, let your own copilots and agents read quality, author rules, and triage failures over MCP.

How it works

Data source ─► Data asset ─► Rules (on the asset) ─┐
                                                    ├─► Rule group ─► Run ─► Results ─► Alerts / Incidents
Data asset ─► Rules ───────────────────────────────┘                          └─► Catalog sync, AI feedback
  1. Connect a data source, warehouse, lakehouse, database, or files. Credentials can stay on your network via the Privacy Agent.
  2. Profile an asset to capture stats, health, and structure.
  3. Author rules directly on the asset, accept AI-suggested expectations or write your own.
  4. Group rules across assets into a rule group, the unit you run and score.
  5. Validate on demand, on a schedule, or from CI/CD and your agents via the API and MCP.
  6. Act, route failures to alerts and incidents, sync results to your catalog, and feed clean signal back to your AI.

Built on GX Core

Measured runs seamlessly on GX Core, the open-source Great Expectations engine:

  • Zero rewrite, existing GX Core expectations work as-is.
  • GX Core API-compatible, point existing pipelines at Measured with a single environment variable.
  • No proprietary lock-in, the open standard your data team already trusts, supercharged with AI, scheduling, RBAC, lineage, and observability.

Key concepts

| Term | Description | |---|---| | Data source | A connection to a warehouse, lakehouse, database, or file store | | Data asset | A specific table or file within a data source, profiled, health-scored, and the place rules live | | Rule | A single assertion about an asset (e.g. "this column is never null") | | Rule group | A runnable set of rules, spanning one or many assets, that carries a health score and produces results | | Result | The output of a run: pass/fail per rule, failure counts, and an AI root-cause summary | | Remediation | Alerts and incidents triggered from results, fire-and-forget or a full incident lifecycle |

Deploy your way

Run Measured as managed cloud or self-host it on your own infrastructure, including air-gapped and VPC-only environments via the Privacy Agent, where data and credentials never leave your network and AI runs against an LLM you control.

Next steps