Elementary

Overview

Elementary is the only genuinely open-source option in data observability. The package and command-line tool are Apache 2.0, verified in the licence file, with no Elastic or business source licence anywhere.

It ships in two editions, and the line between them is sharper than the marketing implies. The open-source edition is a self-maintained tool for teams running dbt; Elementary Cloud is the hosted commercial platform.

Open source gets dbt test collection plus basic freshness and volume anomaly tests, and table-level lineage. The machine-learning monitors, column-level lineage, incident management, governance, single sign-on and certifications are all cloud-only.

Its origin explains its shape: it began by collecting the dbt tests you had already written rather than asking you to rewrite them, and it still integrates dbt artefacts directly into your warehouse.

In the cloud, freshness and volume monitors are genuinely zero-configuration on all production tables, learning update frequency, seasonality and trends from at least 21 days of backfilled metadata.

One claim here is unusual and worth checking against your bill: those monitors read only warehouse metadata, so Elementary states they add no warehouse compute cost. Custom SQL tests, by contrast, do consume compute.

Its privacy model is the strictest here: the cloud needs no permission to read your data, only read access to the Elementary schema, with servers in Frankfurt and an option to store some data in Virginia.

Ella is the agent layer, with agents for test recommendations, triage, governance, cost and cataloguing, and version 2.0 in December 2025 repositioned it as a data and AI control plane that also monitors Python pipelines.

One security note belongs in any evaluation: in April 2026 an attacker published a malicious build of the open-source CLI, remediated within hours. The cloud product and the dbt package were unaffected, but never install that version.

Key features and capabilities

The same headings are used for every data observability entry, so two tools can be read side by side.

What it monitors

• Automated freshness and volume monitors on all production tables in the cloud edition
• dbt test result collection in both editions, including dbt-utils, dbt-expectations and its own package tests
• Data contract tests checking column names, types and nullability against a defined contract
• Custom SQL tests for business rules, plus model and test performance and compute cost metrics
• Python pipeline monitoring arrived with version 2.0; dashboard-level monitors are not published

How incidents are detected

• Machine learning models learn each table's update frequency and forecast whether it is currently fresh
• The model adjusts for update frequency, seasonality and trends, from at least 21 days of backfilled metadata
• Freshness and volume models read warehouse metadata only, so Elementary states they add no compute cost
• Custom SQL tests execute against table data and do consume warehouse resources
• No bad-alert feedback loop or sensitivity scale is published; noise is handled by grouping and ownership routing

Coverage and onboarding

• Open source covers a single dbt project; the cloud covers multiple projects end to end
• The cloud needs only a warehouse connection, and dbt is optional rather than required
• Automated monitors apply to all production tables by default, operational within a few hours of connection
• Bulk test configuration through the interface generates a pull request against your dbt repository
• A test coverage screen with automated recommendations; no sampling policy is published

Triage and root cause

• Incident management is cloud-only; open source gets basic Slack and Teams alerts
• Grouped incidents route context-aware alerts based on ownership and severity
• A triage agent clusters related alerts, analyses causes and proposes or executes fixes
• Data health scores by domain and quality dimension, with critical assets designated
• Ownership assignment on data products, with ticketing integrations in the cloud

Lineage and impact

• Open source gives table-level lineage; column-level lineage to BI is cloud-only
• Lineage is derived from dbt artefacts and the manifest, plus warehouse metadata
• Multi-project lineage shows dependencies across several dbt projects in one graph
• Connecting Looker extends column-level lineage to dashboard level
• Impact analysis is a cloud feature, absent from the open-source edition

Integrations

• More than 14 data platforms, including Snowflake, BigQuery, Redshift, Databricks, Dremio, Trino and Fabric
• dbt Core, dbt Cloud and dbt Fusion in beta, with Airflow and Fivetran for orchestration
• Ten BI tools including Tableau, Looker, Power BI, Sigma, Hex and Metabase
• Code repositories including GitHub, GitLab, Bitbucket and Azure DevOps, needed for pull-request generation
• A CLI, a Python SDK and an MCP server; no general REST API is published

Where it runs and what it costs

• The cloud is software as a service, agentless, collecting metadata and aggregated metrics only
• It needs no permission to read your data, only read access to the Elementary schema
• Servers are in Frankfurt, with an option to store some data in Virginia, plus AWS private links
• The open-source edition is fully self-hosted, with no single sign-on, role-based access, audit logs or certifications
• Priced on editor and viewer seats plus monitored tables, with the AI layer as a credit-based add-on

Pricing

Open sourceFree and Apache 2.0; cloud is quote-only

The open-source edition is free under Apache 2.0 with no seat or table limits, which makes it the genuine free option in this category, though its feature set is narrower than the cloud's. Elementary Cloud has three tiers, Scale, Enterprise and Unlimited, all quote-only with no figures: ten editor seats and a thousand tables at the bottom, rising to unlimited seats and three thousand tables, each charging for extra tables. The AI layer is a credit-based add-on. A 30-day trial offers unlimited seats, environments and tables.

Vendor pricing page →

Demos and videos

• Open-source data observability hands-on tutorial Tutorial · YouTube
• Test coverage on Elementary Cloud Product demo · YouTube

About Elementary

Elementary was founded in August 2021 by Maayan Salom, its chief executive, and Or Avidov, its chief product and technology officer, and operates as Elementary Data Inc.; it publishes no headquarters. Its handbook records a seed round in April 2022 but publishes no amount, no investors and no later round, which is a notable gap against its peers. It reports more than 1,500 data teams and a 3,000-strong community, naming Patagonia, Fiverr, Zoom, Spotify and Elastic. The open-source project is Apache 2.0 with around 92 contributors and remains actively maintained, releasing through September 2026.

Founded 2021 · elementary-data.com