Datafold

Overview

Datafold answers one question: if I merge this change, what happens to the data? Its core mechanism is a data diff, a value-level comparison between two versions of a table, row by row and column by column, within one database or across two.

The everyday use is regression testing on a pull request. Your CI builds the changed models into a staging schema, Datafold compares that against production and posts the differences as a comment on the pull request, including which rows changed and by how much.

It pairs that with column-level lineage built from warehouse query logs, dbt metadata and BI tools, so a change can be traced to the dashboards and reverse-ETL jobs it would affect. That combination, a diff plus impact, is what distinguishes it from ordinary testing.

Cross-database diffing makes it a migration tool as well: it validates that a rewritten pipeline or a move to a new warehouse produces the same values, and a migration agent translates legacy SQL and checks the result.

One important point for anyone who remembers the open-source version: data-diff was archived in May 2024, and Datafold said plainly it would no longer support or develop it. Everything current is the commercial product.

It has no runner of its own; it plugs into GitHub, GitLab or Bitbucket CI. Deployment is multi-tenant SaaS in Oregon or Ireland, or a single-tenant VPC on AWS, GCP or Azure.

Key features and capabilities

The same headings are used for every dataops & ci/cd entry, so two tools can be read side by side.

What it does for data teams

• Catches breaking data changes before merge by diffing a staging build against production
• Validates migrations and replication with cross-database diffs
• Flags downstream dashboards and reverse-ETL jobs a change would affect
• Does not run dbt or host pipelines; it plugs into the CI that does

Pipelines and automation

• No runner, workflow language or agents of its own
• Invoked from GitHub, GitLab or Bitbucket CI, or through the REST API
• Requires a CI step that builds the changed models into a staging schema first
• Monitors can be defined as version-controlled YAML

Testing and change checks

• Value-level diffs, row by row and column by column, in or across databases
• AI code review of SQL and pipeline code, usable with or without diffs
• Column-level impact analysis on downstream assets
• Monitors for diffs, metrics, tests and schema changes, with alerting

Environments and isolation

• Works against a staging schema built from the branch; it does not create environments
• No data branching or zero-copy environments
• Diff results can be browsed, exported as JSON or written back to a database

Approvals, secrets and audit

• Role-based access control, and single sign-on with Okta, Google or SAML
• Service accounts as machine identities for CI
• Approvals and protected branches are enforced by the Git platform, not Datafold
• A dedicated audit log is not published

Integrations

• Warehouses: Snowflake, BigQuery, Databricks, Redshift, PostgreSQL, SQL Server, Trino and more
• dbt Core and dbt Cloud, plus Airflow and other orchestrators
• GitHub, GitLab and Bitbucket
• A Slack bot for questions about data, and MCP for coding agents

How it runs

• Multi-tenant SaaS in Oregon or Ireland, with private link, VPC peering and IP allowlists
• Single-tenant VPC on AWS, GCP or Azure, in three management models
• Dedicated deployments run on Kubernetes
• No on-premises option is published

Pricing

Price on request

Quote-only: the pricing page redirects to a contact form, with no tiers, prices or free-tier limits published. The only figures the company ever published were in a 2022 blog post, which described a free tier and a cloud tier from $799 a month, and should be treated as historical. A trial is offered but its length is not stated.

Vendor pricing page →

Demos and videos

• Fully automated dbt-to-dbt migrations with Datafold Product demo · YouTube
• Datafold's migration agent Product demo · YouTube

About Datafold

Datafold was founded in 2020 by Gleb Mezhanskiy, who has said the idea came from a two-line SQL change at Lyft that corrupted hundreds of tables and dashboards. The company is all-remote and does not publish a head office. It has raised $22m in total, backed by Y Combinator, NEA and Amplify Partners, including a $20m Series A in November 2021. The product is proprietary SaaS; the open-source data-diff project it was once known for was MIT licensed and archived in 2024. It publishes SOC 2 and HIPAA compliance.

Founded 2020 · datafold.com