Earlier this year, we launched agent mode in Dolt Workbench. It’s a Cursor-like agentic chat interface that allows you to safely interact with your databases.
Originally, agent mode only supported Dolt, MySQL, and Postgres databases. Today, we’re announcing that we’ve also added support for Doltgres. In this article, I’ll walk through an example showing how to use agent mode for a data import task.
Agent Data Import Example#
To get started, you’ll need a running doltgres server and a copy of Dolt Workbench downloaded on your computer. For this example, I’m starting with a completely empty database. After connecting to the doltgres server from the workbench, click the orange robot icon at the top right of the screen to open the agent panel. If it’s your first time using agent mode, you’ll be prompted to supply an Anthropic API key.
Now, with an empty database and a fresh agent session, it’s time to do something interesting. I wanted to see how well the agent could handle a somewhat ambiguous data import task. To test this out, I asked the agent to load in Pagila, the canonical Postgres example database. This was my prompt:
I was intentionally vague here. That link points to a GitHub repository with a bunch of scattered .sql files. I wanted to see if the agent could accurately decipher the schema for each of the tables, pull down the data appropriately, and load everything into my doltgres database. Additionally, I knew that there would likely be features of this database that are not yet fully supported in doltgres. I was curious to see how the agent would adapt when it ran into these issues.
About 10 minutes and 20 tool calls later, I was greeted with this message:
When using agent mode in the workbench, the agent will never make direct commits without your permission. This gives you the luxury of examining the diff before any changes become part of your commit history. Immediately, I could see in the main view that all 15 tables had been created:
And here’s a summary of the diff:
Before committing, I also wanted to know what changes were necessary in order to make this database fully Doltgres-compatible.
After confirming with the team that this was all in accordance with the current state of Doltgres, I allowed the agent to commit my changes. With this done, I had a fully version-controlled Pagila clone!
Conclusion#
This is just one example of a task you can use agent mode for with a Dolt or Doltgres database. For a more complete summary of its capabilities, check out this overview blog. For an example of a more complex data analysis task, read about how we used agent mode to uncover Medicaid fraud. This is a feature we think has a ton of potential, so we’d love for you let us know what you think on Discord.