Your AI is only as good as the data beneath it.
Measured is the data quality foundation for everything you build with AI. It continuously validates the data going into your models and the results coming out: the quality layer your models, agents, and RAG pipelines run on.
Your data is ready for AI
Measured checked 10,200 records, caught 33 with negative values, and opened an incident, before any of it reached your model.
Connects to your whole stack
Measure data wherever it lives
Plus lakehouse, files, and more. Need one we don't list? Ask us.
AI doesn't fail loudly. It fails quietly, on bad data.
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. Garbage in isn't just garbage out anymore: it's hallucinations, bad decisions, and eroded trust at scale.
Bad inputs poison fine-tunes and retrieval, confidently and invisibly.
Distributions shift under your models with no alarm until accuracy tanks.
Generated records flow downstream unchecked, compounding errors.
One quality layer for everything your AI touches.
Quality in
Validate every input your models depend on: training sets, feature tables, RAG sources. Catch nulls, drift, schema breaks, and referential gaps before they reach the model.
Quality out
Treat AI results as first-class data assets. Validate generated records, scores, and the pipelines downstream of your models so errors never compound silently.
Your LLMs, plugged in
A native MCP server lets your own LLMs and agents query Measured, author rules, and investigate failures in plain language. AI that manages data quality for AI.
AI-authored rules
Profile any asset and Measured drafts the right checks: completeness, accuracy, uniqueness, ranges, consistency, structure.
Anomaly detection
Statistical and LLM-driven drift detection on volume, schema, and distributions, with plain-English explanations.
MCP server
Point Claude, your copilots, or in-house agents at Measured. They read quality, write rules, and triage failures over MCP.
Ask AI, in context
A contextual assistant that knows your org, workspace, and team, and links to the docs so nobody has to read them.
Bring your own LLM
OpenAI, Anthropic, Ollama, or any custom endpoint. Run it in the cloud or fully on-prem via the Privacy Agent.
Explainable failures
Every failed check comes with an AI root-cause summary and a suggested fix, not just a red X.
From raw table to trusted AI input, in five steps.
Connect
Point Measured at your warehouse, lakehouse, or database. Credentials can stay on your network via the Privacy Agent.
Profile
Auto-profile assets for stats, health, and structure: the raw signal the AI uses to understand your data.
Let AI draft rules
Accept AI-suggested checks on the asset, or write your own. Group them into cross-asset rule groups.
Validate
Run on demand, on a schedule, or as a step in your pipeline: Airflow, GitHub Actions, dbt, or any orchestrator via the API. Every run scores asset health and can fail the job.
Act
Route failures to alerts and incidents, sync results to your catalog, and feed clean signal back to your AI.
Validate where the data moves, not after it lands.
Measured is a step in the pipelines you already run. Call it from your orchestrator, your CI, or your dbt project: it validates the data, scores it, and fails the task when quality doesn't meet the bar, so bad data stops before anything downstream, models included, ever sees it.
Orchestrators
Apache Airflow gets a first-class provider: an operator that runs a validation and fails the task if it fails, a sensor for long runs, and the full result in XCom. Dagster, Prefect, Cloud Composer, MWAA, and Databricks Workflows use the same three-call pattern over the API or Python SDK.
CI/CD
The GitHub Action runs the validations for the dbt models a pull request touches against the PR's own staging schema and posts a Check Run. Make it a required check and bad data can't merge. GitLab, Jenkins, and CircleCI: one curl.
dbt
Ingest manifest.json and run_results.json and every model becomes a data asset with its last run's outcomes attached. Add the dbt Cloud webhook and production runs fire their validations automatically.
from dq_cloud_airflow.operators import DQCloudCheckpointOperator
validate_orders = DQCloudCheckpointOperator(
task_id="validate_orders",
org_id=ORG, workspace_id=WORKSPACE,
checkpoint_id=ORDERS_VALIDATION, # fails the task on a failed run
)
load_orders >> validate_orders >> publish_orders- uses: Measured-Cloud/Measured-Cloud/providers/github-action@main
with:
dq-cloud-url: https://app.measured.cloud
dq-cloud-api-key: ${{ secrets.MEASURED_API_KEY }}
staging-schema: dbt_pr_${{ github.event.pull_request.number }}
checkpoint-tag-filter: ci-blockerAnything else: trigger, poll, gate. Three REST calls or pip install dq-cloud. See the pipeline recipes
Runs the engine your data team already trusts.
Measured runs seamlessly on Great Expectations Core, the open-source data validation engine. Your existing checks work as-is, no rewrite, no lock-in, now supercharged with AI authoring, scheduling, RBAC, lineage, and observability.
- 100+ open-source check types, plus custom SQL
- API-compatible: point existing pipelines at Measured with one environment variable
- No proprietary rule format to get trapped in
# one environment variable, no code changes
export GX_CLOUD_BASE_URL=https://app.measured.cloud
export GX_CLOUD_ACCESS_TOKEN=dqk_•••••••••
# your existing pipeline just works
context = gx.get_context(mode="cloud")
context.checkpoints.get("portfolio-quality").run(){
"mcpServers": {
"measured": {
"type": "http",
"url": "https://mcp.measured.cloud/mcp",
"headers": { "Authorization": "Bearer dqk_•••••••••" }
}
}
}claude> Ran it — 9,940 of 10,200 records passed. Two rules failed on portfolio_weight_pct…
Bring Measured to the LLM you already use.
Measured ships a native Model Context Protocol server — a hosted endpoint plus a personal API key. Point Claude Desktop, Claude Code, Cursor, or your in-house agents at it and your data-quality workflow runs in plain language, from the tool you already work in.
- See sources, assets, schemas, profiles, and health
- Draft and update rules from real profile data — you approve before anything saves
- Trigger validations, poll runs, and get plain-English pass/fail summaries
- Triage failures and anomalies, create alerts and schedules
- Ask product questions — the docs are searchable over MCP too
37 tools, one bearer key that acts as you — same roles, same workspace isolation, revocable in one click. On Enterprise it all runs against your own on-prem LLM via the Privacy Agent.
Infrastructure you can put under production AI.
Privacy Agent
Run validation on-prem, between your data sources and the cloud. Data, credentials, and results never leave your network.
12+ connectors
Snowflake, BigQuery, Databricks, Redshift, Postgres, Oracle, SAP HANA, ClickHouse, Trino, MS SQL, MySQL, lakehouse.
Organization & teams
Six-role RBAC, SSO via Okta/Entra/Google, restricted PII/PHI/PCI assets, and full tenant isolation.
Alerts & incidents
Email, Slack, Teams, PagerDuty, Opsgenie, ServiceNow, Jira, webhook, plus a built-in incident lifecycle.
Catalog integrations
First-party push to DataHub, Atlan, Alation, and Data.world so quality shows up where people already look.
API-first
Every action is an API call. Automate from CI/CD, orchestrators, and your own agents, and pull metrics into PowerBI or Tableau. First-class Airflow provider and GitHub Action included.
Start free. Scale per seat.
For small teams getting started
- 1 user, 1 workspace
- 2 data sources, up to 25 rules
- AI rule suggestions
- Manual & scheduled validation
The full AI quality platform for your team
- Unlimited data sources & rules
- Full AI toolkit + MCP server
- Unlimited connectors
- Rule groups, alerts & incidents
- Organization & teams
Run it as production infrastructure
- Everything in Team
- Self-host & Privacy Agent
- On-prem LLM, SSO/SAML, RBAC
- Audit, SLA & dedicated support