dbt-autodoc 1.0.12

Automated documentation generator for dbt projects using Google Gemini AI

DBT Autodoc Documentation

dbt-autodoc is the ultimate tool for Automated Documentation and Logging for your dbt projects. It combines the power of Google Gemini AI with a robust Database Logging system to ensure your documentation is always up-to-date, accurate, and auditable.

🌟 Why dbt-autodoc?

• 🤖 Automatic AI Documentation: Generate comprehensive descriptions for your tables and columns automatically.
• 💾 Database Logging & History: Every description is stored in a database (duckdb or postgres). This acts as a "Source of Truth" and provides a full history of changes.
• 🔄 Full Synchronization: Seamlessly integrates with dbt-osmosis to keep your YAML files in sync with your SQL models.
• 🔒 Protect Manual Work: Respects human-written documentation. If you write it, we lock it.
• 👥 Team Ready: Use Postgres to share documentation cache across your entire team.

🛠️ Setup

1. Install:

   pip install dbt-autodoc
   

2. Configuration: Run dbt-autodoc to generate dbt-autodoc.yml. Important: Edit company_context in this file to give the AI knowledge about your business logic.

3. Environment Variables:

   GEMINI_API_KEY=your_api_key_here
   POSTGRES_URL=postgresql://user:pass@host:port/db (optional)
   

📋 Recommended Workflow

For the best results, follow this step-by-step workflow to ensure accuracy and control:

1. Preparation: Update your dbt project and context.

   dbt run
   # Edit dbt-autodoc.yml with company_context
   

2. Sync Structure (No AI): Regenerate YAML files to match the SQL models. This ensures all new columns are present.

   dbt-autodoc --regenerate-yml
   

3. Generate Table Descriptions (SQL): Generate AI descriptions for your models (tables/views).

   dbt-autodoc --generate-docs-config-ai --model-path models/staging
   

4. Manual Review (Important): Open your YAML files. Review the structure and any existing descriptions. If you manually update a description here, it will be protected from AI overwrites in the next step.

5. Generate Column Descriptions (YAML): Use AI to fill in the missing column descriptions.

   dbt-autodoc --generate-docs-yml-ai --model-path models/staging
   

6. Propagate & Save: Run osmosis again to apply inheritance rules to all the dbt project, then run the tool again to save the final state (including inherited descriptions) to the database.

   dbt-autodoc --regenerate-yml
   dbt-autodoc --generate-docs-yml-ai --model-path models/staging
   

7. Next Layer: Repeat steps 2-6 for models/intermediate, models/marts, etc.

🚀 Quick Start (Automated)

If you trust the process and just want to run everything at once:

dbt-autodoc --generate-docs-ai

🧠 How the AI Works

When generating a description for a column or table, the AI considers multiple inputs to produce the most accurate result:

1. Company Context: The high-level business logic defined in your config.
2. Model SQL: The actual code of the model being documented.
3. Existing Descriptions: Any existing documentation or comments in the file.
4. Upstream Logic: (Implicitly via Osmosis inheritance) Context from upstream models.

It synthesizes all these inputs to write a concise, technical description.

📖 Arguments Reference

Argument	Description
--regenerate-yml	Structure Only. Only runs dbt-osmosis to regenerate YAML files from dbt models. Does not sync to DB or call AI.
--generate-docs-ai	🔥 Full Auto. Runs the complete workflow: SQL generation, Osmosis sync, and YAML generation using AI.
--generate-docs	🔄 Full Sync. Runs the complete workflow using only the database cache (no AI).
--model-path	Restrict processing to a specific directory (e.g. models/staging).
--generate-docs-config-ai	Generate table descriptions in .sql files using AI.
--generate-docs-yml-ai	Generate column descriptions in .yml files using AI.
--generate-docs-config	Sync .sql files from cache (no AI).
--generate-docs-yml	Sync .yml files from cache (no AI).
--cleanup-db	Reset Database. Wipes the description cache and history.
--concurrency	Max threads for AI/DB requests (default: 10).

📄 License

MIT License - see LICENSE for details.

🙏 Attribution

Brought to you by JustDataPlease.