Buster is a new framework for AI-powered self-serve
We believe that the future of AI analytics is about helping data engineers build powerful, self-serve experiences for their users. Below is an overview of how we help you do this.
Purpose-built for dbt
Buster is purpose-built for dbt
Feeding an LLM a raw warehouse schema will result in a very confused LLM. Or worse, a confident LLM that makes incorrect assumptions.
This is why we specifically built Buster for teams that use dbt.
Buster is intentionally built to operate within the scope of your dbt models & documentation. This does a couple of really important things:
It allows you to document all of your business logic & tribal knowledge into code
It creates a single source of truth
It reduces the assumptions that AI agents need to make
It allows us to validate queries & deterministically identify when an AI agent makes assumptions (e.g., does things that are not defined in your models)
It allows you to continuously improve your AI agents & define nuance over time
Code-based & Git-native
Everything in Buster is code-based
Everything in Buster is code-based and lives (as files) in your own Github Repo. This enables you to manage everything from your CLI & CI/CD pipeline.
More importantly, it enables things like:
Version control
Support for multiple environments
Identifying discrepancies or duplicative content
Auto-generating pull requests for suggested model improvements
Generating bulk changes across your repo
Automatically fixing impacted dependencies when there are breaking changes
And much, much more (we're building some cool stuff)
Enrich your dbt models
Enrich your dbt models with AI-generated documentation
With our CLI tool, you can auto-enrich your dbt models with additional metadata & documentation. The CLI tool can do things like:
Scan your source directory for SQL files & existing documentation
Assess dbt lineage, ingest & organize metadata from your BI tools, etc
Automatically create YAML documentation files
Generate robust documentation with custom metadata fields (e.g., model descriptions, column descriptions, enum indexing, data types, predicted joins, etc)
Document unique business context & terminology
Changes in your dbt repo will automatically be reflected in Buster.
Implement guardrails & AI safety
Implement guardrails & control AI querying capabilities
Buster is trained to strictly work within the bounds of the dbt models that have been granted query access (by you). Anytime Buster attempts to generate a query that is not explicitly defined in the underlying data model, the query gets flagged.
We do this by running a set of rigorous tests every time Buster generates SQL or Python. These evaluations detect lots of things, but the most important are:
Did the AI have to make assumptions about things that are not explicitly defined in the data model(s) or data catalog?
If assumptions were made, how severe was each individual assumption?
You can do all kinds of automated things with flagged assumptions:
You can allow the AI data analyst to run the query & notify the end user of the assumptions that were made.
You can block the query from being run & return an error message to the end user.
You can send a ticket to Jira or Linear for review.
You can have Buster generate a pull request with a suggested model improvement or documentation update.
Self-improving data models
Improve data models with auto-generated pull requests
Buster was intentionally built to create strong feedback loops between the BI layer (the AI data analyst) and the modeling layer. This allows Buster to constantly be optimizing your dbt models & documentation files.
identify a potential model improvement → create a new branch → generate an update to your data model → send you a pull request
You can then review the request & merge it with one click.
Open source
Open source and deployable anywhere
Buster is fully open source. No vendor lock-in. No surprises.
Ready to start leveraging AI data agents at your org?