Skip to main content
Buster is a platform for building autonomous AI agents that automate dbt and data engineering workflows. Agents are YAML configuration files you commit to your repository that describe tasks in natural language—like updating documentation, reviewing code, or adapting to schema changes. Agents execute automatically in response to triggers like pull requests, schedules, or Airflow pipeline failures.
Data engineering with dbt involves repetitive tasks that require context about your data and code: keeping documentation current as models change, reviewing PRs for best practices and breaking changes, adapting staging models when upstream schemas change, and monitoring data quality. These tasks are tedious to do manually but perfect for AI agents with access to your warehouse and repository.
Traditional scripts are procedural—you write explicit code defining every step. Buster agents are declarative—you describe what you want accomplished in natural language, and the agent figures out how to do it. Agents can read your code, query your warehouse, reason about patterns, and adapt to different scenarios without requiring you to handle every edge case.
Safety is built into Buster’s design:
  • Sandboxed execution: Agents run in isolated Daytona sandboxes with no general internet access
  • Network isolation: Agents can only access your repository, warehouse, and configured integrations (GitHub, Slack)
  • Complete audit logs: Every agent action is logged—files read, queries run, decisions made
  • GitHub Check Runs: Every execution creates a check run so you can see exactly what happened
  • Version control: Agent configs are committed to git, so you can review changes and roll back
  • Validate before deploy: Use buster deploy --dry-run to validate configuration before deploying
You need:
  • A GitHub account with admin access to your repository
  • A data warehouse (Snowflake, BigQuery, Redshift, Databricks, PostgreSQL, MySQL, ClickHouse, SQL Server, or Supabase)
  • Warehouse credentials (read-only access is sufficient)
Setup takes about 10 minutes: connect GitHub, connect your warehouse, create a buster.yml, and deploy your first agent.
No. Agents are configured using simple YAML files with natural language instructions. If you can write dbt YAML and describe a task in plain English, you can create agents. No Python, no complex scripting—just describe what you want done.
Yes. Use the CLI to validate your configuration:
buster deploy --dry-run --verbose
This checks your buster.yml and agent files for errors without actually deploying. Once deployed, test agents by creating a sample PR or waiting for the next scheduled run, then review the execution in the Runs tab.
Buster integrates with:
  • Snowflake
  • BigQuery
  • Amazon Redshift
  • Databricks
  • PostgreSQL
  • MySQL
  • ClickHouse
  • SQL Server
  • Supabase
Agents connect with read-only credentials by default.
For pricing details, visit buster.so or contact [email protected].
Common use cases include:
  • Documentation: Auto-update model and column descriptions when code changes
  • Code review: Check PRs for SQL anti-patterns, naming conventions, and missing tests
  • Schema changes: Detect upstream schema changes and adapt staging models automatically
  • Data quality: Monitor freshness, null rates, referential integrity, and anomalies
  • Pipeline failures: Investigate and report on Airflow DAG failures
  • Audits: Regular checks for documentation coverage, unused models, or compliance issues
See our Examples section for real agent configurations.
When you connect your repository and warehouse, Buster automatically:
  1. Creates an AGENTS.md file — A comprehensive guide capturing your tech stack, conventions, modeling philosophy, and domain knowledge
  2. Profiles your models — Runs metadata queries, analyzes columns, discovers patterns, and generates documentation
This context is loaded into every agent execution, giving them deep understanding of your specific project—not just generic dbt knowledge.
Agents run in secure, isolated Daytona sandboxes—cloud environments managed by Buster. Each agent execution:
  1. Provisions a fresh sandbox with your repository cloned
  2. Injects warehouse credentials and GitHub tokens securely
  3. Loads your AGENTS.md and trigger context (PR details, etc.)
  4. Executes with access to files, SQL, and bash commands
  5. Takes actions like creating PRs, posting comments, or sending Slack messages
  6. Reports results via GitHub Check Runs and logs everything
Agents never have persistent access—each run is isolated and stateless.
Check these resources:For additional support, email [email protected].