Relium

Product Introduction

1. Definition: Relium is a pre-merge reliability and semantic verification layer for SQL and dbt (data build tool) pipelines. It operates as a GitHub App integrated directly into the CI/CD (Continuous Integration/Continuous Deployment) workflow for data teams.
2. Core Value Proposition: Relium exists to prevent silent, logic-based data failures from reaching production. Unlike post-deployment data observability or testing tools, it performs semantic analysis and downstream impact tracing on pull requests (PRs) to provide a data-driven release decision (ALLOW, WARN, BLOCK) before code is merged, ensuring data reliability and semantic consistency.

Main Features

1. Semantic Change Analysis: Relium performs a deep semantic diff on SQL and dbt model changes, going beyond syntax. It identifies when logical operations like JOIN types effectively change (e.g., a filtered LEFT JOIN becoming an inner join) or when metric definitions are subtly altered. This analysis is based on the compiled dbt manifest artifacts from both the base and head commits.
2. Downstream Impact Tracing: The tool automatically traces the full lineage impact of a proposed change. Using the dbt project's manifest (which includes models, tests, exposures, and contracts), Relium maps out all downstream models, metrics, and dashboards (exposures) that would be affected, providing a clear visualization of potential blast radius.
3. Evidence-Based Release Gating: Relium consolidates analysis into a single, enforceable release decision. It evaluates semantic risk, downstream exposure, and optional warehouse evidence (like row count changes or null rate shifts) to output an ALLOW, WARN, or BLOCK verdict. This decision is published directly on the GitHub pull request as a check run and comment, integrating seamlessly into existing developer workflows.

Problems Solved

1. Pain Point: The "silent logic failure" in data pipelines. Traditional dbt tests (uniqueness, not-null, schema) validate data shape but not business logic meaning. A change can pass all tests, deploy cleanly, and still corrupt key metrics (e.g., revenue, customer count) without immediate alerts, leading to costly decision-making based on faulty data.
2. Target Audience: Data engineers, analytics engineers, and data platform teams responsible for maintaining reliable dbt projects and SQL-based data transformations. It is also critical for data governance roles and business intelligence consumers who depend on consistent metric definitions.
3. Use Cases: Essential for reviewing any change to core fact or dimension models, refactoring SQL logic, modifying JOIN conditions, adjusting filter placements, and updating metric calculations within dbt. It is particularly valuable in regulated industries or organizations where data accuracy is paramount for financial or operational reporting.

Unique Advantages

1. Differentiation: Relium is fundamentally proactive (pre-merge) rather than reactive (post-incident). It contrasts with data observability platforms (like Monte Carlo, Datafold) that primarily alert after a break occurs. While tools like Datafold offer data diffing, Relium focuses on semantic diffing and automated release gating as a integrated check, not just a diff report for manual review.
2. Key Innovation: The synthesis of compiled dbt lineage, semantic analysis, and optional warehouse evidence into an automated, policy-driven release gate. The concept of a "reliability layer" that outputs a clear ALLOW/WARN/BLOCK decision—akin to a security or code quality scan—specifically for data logic changes is a novel approach in the modern data stack.

Frequently Asked Questions (FAQ)

1. How does Relium work with dbt Cloud and dbt Core? Relium works with both dbt Core and dbt Cloud by consuming the compiled manifest.json artifact. Teams configure a CI job (e.g., in GitHub Actions) to compile the dbt project for both the base and head commits of a pull request and submit these manifests to Relium's API for analysis.
2. Does Relium access or store my raw production data? No. The core analysis uses only your code repository and dbt manifest artifacts. For enhanced evidence, an optional collector runs within your own environment using read-only warehouse credentials. This collector submits only aggregate metadata (e.g., row counts, null percentages) and never accesses or transmits individual row-level data.
3. Can Relium block merges automatically to enforce data quality? Yes, on the Pro plan. Relium can be configured to fail the GitHub status check on a BLOCK decision, which can prevent merging based on your branch protection rules. This enables teams to enforce data reliability standards directly in their development workflow.
4. What data warehouses and SQL dialects does Relium support? The semantic analysis and lineage tracing are warehouse-agnostic, working with any dbt project (BigQuery, Snowflake, Redshift, etc.). The optional warehouse evidence collector currently supports PostgreSQL and SQLite, with more adapters planned.
5. How is Relium different from writing more dbt tests? dbt tests are essential for validating data integrity constraints (e.g., uniqueness, referential integrity). Relium operates at a higher level, verifying that the business logic and semantic meaning of your models and metrics remain consistent. It answers the question, "Did the definition of this key metric change?" which is outside the scope of standard data testing.