# Dolt: Git-style version control for SQL databases [Dolt](https://www.dolthub.com/) is a **MySQL-compatible SQL database with Git semantics built into its core**. The same workflow used for code (branch, commit, diff, merge, log) applies directly to database tables. It is a drop-in replacement for MySQL, so any application or tool that connects to MySQL can use Dolt without code changes. ![Dolt homepage highlighting Git-like versioning, SQL interface, drop-in replacement, instantaneous rollbacks, and collaboration tools](https://imagedelivery.net/xZXo0QFi-1_4Zimer-T0XQ/edc610dd-8082-472e-42ae-07b4c3b2d300/md1x =1280x720) ## The problem with existing approaches **Standard relational databases** (MySQL, PostgreSQL) provide schema enforcement, constraints, and powerful querying, but changes are immediate and often irreversible. There is no built-in diff, no commit history without a custom audit log, and no clean rollback mechanism. **Data in Git** (CSV, JSON, YAML files) provides version control with commits and pull requests, but sacrifices all database capabilities. There are no primary keys, no type enforcement, querying requires loading and parsing files, and diffs on CSV files are line-level rather than cell-level, making them hard to review and prone to merge conflicts. Dolt provides both: a relational database with full SQL and a built-in Git workflow. ## How it works: Prolly Trees Traditional databases use B-Trees optimized for reads and writes but not designed for efficient versioning. Dolt uses Prolly Trees (Probabilistic B-Trees), a content-addressed, immutable data structure where each node is identified by a hash of its contents. ![Diagram from Dolt documentation showing the structure of a Prolly Tree with a root node and internal nodes that hash their contents](https://imagedelivery.net/xZXo0QFi-1_4Zimer-T0XQ/ee2547a2-8eda-40be-40c9-4df8dc52db00/public =1280x720) When data changes, Dolt does not modify the existing tree. It creates new parent nodes pointing to the changed data while reusing pointers to unchanged nodes. This means a commit only stores the new data and new pointers, making branches, diffs, and commits nearly instantaneous regardless of database size. ## Setup Install the Dolt CLI following the [official installation documentation](https://docs.dolthub.com/introduction/installation) for macOS, Windows, or Linux. Initialize a repository in a new directory: ```command mkdir dolt-tutorial && cd dolt-tutorial ``` ```command dolt init ``` This creates a `.dolt` subdirectory for history, configuration, and metadata. ## Creating tables and inserting data Start the interactive SQL shell: ```command dolt sql ``` ![Terminal showing the DoltSQL shell welcome message indicating statements must be terminated with a semicolon](https://imagedelivery.net/xZXo0QFi-1_4Zimer-T0XQ/9eb69f46-a8a1-4e0f-a63f-bf75d302b800/public =1280x720) ```sql CREATE TABLE employees ( id INT NOT NULL, name VARCHAR(255), title VARCHAR(255), PRIMARY KEY (id) ); INSERT INTO employees (id, name, title) VALUES (1, 'Alice Johnson', 'Software Engineer'), (2, 'Bob Williams', 'Product Manager'), (3, 'Charlie Brown', 'Data Analyst'); ``` Update a record: ```sql UPDATE employees SET title = 'Senior Data Analyst' WHERE id = 3; ``` Exit the shell: ```sql exit; ``` ## Tracking changes Check what has changed: ```command dolt status ``` View a structured cell-level diff: ```command dolt diff ``` ![Color-coded diff in the terminal showing the old row with a minus sign and new row with a plus sign highlighting the exact cell that changed](https://imagedelivery.net/xZXo0QFi-1_4Zimer-T0XQ/2b02c6c9-5a47-4cb5-ffb2-60f6f65f0800/lg1x =1280x720) The diff shows which rows and columns changed, not just which lines of text differ. Stage and commit: ```command dolt add employees ``` ```command dolt commit -m "Promote Charlie Brown to Senior Data Analyst" ``` ## Branching and merging Create and check out a new branch: ```command dolt checkout -b add-feature-flags ``` Make changes on the branch using the `-q` flag for single queries: ```command dolt sql -q "CREATE TABLE feature_flags (flag_name VARCHAR(100) PRIMARY KEY, is_enabled BOOLEAN);" ``` ```command dolt sql -q "INSERT INTO feature_flags (flag_name, is_enabled) VALUES ('new-dashboard', false);" ``` ```command dolt add feature_flags ``` ```command dolt commit -m "Add feature_flags table for new dashboard" ``` Merge back into `main`: ```command dolt checkout main ``` ```command dolt merge add-feature-flags ``` ## Audit log ```command dolt log ``` This displays every commit with hash, author, date, and message. The full history is also queryable via system tables such as `dolt_log`, `dolt_diff_`, and `dolt_commits`, allowing SQL queries over commit history. ## Running as a MySQL server ```command dolt sql-server ``` This starts a MySQL-compatible server that any ORM, BI tool, or application can connect to on the default port. The version control commands remain available via the CLI while the server runs. ## DoltHub [DoltHub](https://www.dolthub.com/) is a hosting platform for Dolt databases, analogous to GitHub for code repositories. It supports push, pull, clone, and pull requests for data changes, enabling team-based data review workflows. ## Final thoughts Dolt is most valuable for data that changes over time and needs to be **audited, reviewed, or rolled back: application configuration, feature flags, curated reference datasets, and data pipelines** where a bad import needs to be undone cleanly. For standard transactional application data where version history is not a requirement, a conventional database remains simpler. The Prolly Tree implementation means that branching and diffing do not get slower as the database grows, which distinguishes it from approaches that implement versioning through audit triggers or shadow tables. Source code and documentation are at [github.com/dolthub/dolt](https://github.com/dolthub/dolt).