Introducing Consort: Test-driven development on a branching database

Databricks Field Engineering has released Consort, an open-source agentic framework designed to integrate test-driven development (TDD) with branching databases. The tool addresses a long-standing gap where databases, unlike code or infrastructure, could not be branched or tested in isolation, forcing developers to rely on mocks or shared staging environments. By leveraging Lakebase Postgres, Consort allows engineers to create isolated database branches in constant time, enabling destructive tests and schema changes without impacting team environments.

Consort structures development workflows into two lanes: a spec-first design lane for defining intent and a build lane for executing the full TDD cycle against a live database branch. Each role on the team—such as product owner, architect, or DBA—acts as an agent guided by a conductor, with scoped context packages provided to optimize performance. The framework supports parallel experiments on isolated branches, allowing teams to explore multiple approaches before selecting the optimal solution.

The framework includes a VS Code plugin for unified diff views of Git and Lakebase branches, alongside an observability dashboard that tracks each development turn in real time. These tools provide transparency into the process, enabling engineers to inspect artifacts, prompts, and working software at each gate before approval. Consort is designed to be adopted incrementally, requiring only Lakebase Postgres during development while supporting Java, Python, or Node for the final application.

Consort is open-source and community-supported, with no SLA requirements or platform bundling. It can be deployed using the /consort:start command, with a StockFlow example provided to demonstrate its capabilities. The project invites contributors to collaborate on GitHub, where the repository is hosted for those interested in exploring its internals or contributing to its development.