If you're a dbt user, you know the power of well-crafted data models. You've invested time building clean, tested, and documented transformations that your team relies on. Your dbt models represent the single source of truth for analytics, reporting, and increasingly—machine learning features.

But here's the challenge: when your ML team wants to use these models for production predictions, they often need to rebuild the same transformations in their feature store. Your beautiful dbt models, with all their logic and documentation, end up getting reimplemented elsewhere. This feels like wasted effort, and it is.

You've already done the hard work with dbt:

- **Transformed raw data** into clean, aggregated tables

- **Documented your models** with column descriptions and metadata

- **Tested your logic** to ensure data quality

- **Organized your transformations** into a maintainable codebase

These models are perfect for machine learning features. The aggregations you've built for your daily reports? Those are features. The customer attributes you've enriched? Features. The time-based calculations you've perfected? You guessed it—features.

The problem isn't your models—they're great. The problem is getting them into a system that can serve them for real-time ML predictions with low latency and point-in-time correctness.

Feast's dbt integration is designed with one principle in mind: **your dbt models should be the single source of truth**. Instead of asking you to rewrite your transformations, Feast reads your dbt project and automatically generates everything needed to serve those models for ML predictions.

Here's how it works:

1. **Tag your dbt models** that you want to use as features (just add `tags: ['feast']` to your config)

2. **Run `feast dbt import`** to automatically generate Feast definitions from your dbt metadata

3. **Deploy to production** using Feast's feature serving infrastructure

- Column names, types, and descriptions from your schema files

- Table locations from your dbt models

- All the metadata you've already documented

Then it generates Python code defining Feast entities, data sources, and feature views—all matching your dbt models exactly. Your documentation becomes feature documentation. Your data types become feature types. Your models become production-ready features.

The best part? **You don't change your dbt workflow at all.** Keep building models the way you always have. The integration simply creates a bridge from your dbt project to production ML serving.

Let's walk through a real example. Imagine you're a data engineer at a ride-sharing company, and you've already built dbt models to track driver performance. Your analytics team loves these models, and now your ML team wants to use them to predict which drivers are likely to accept rides.

Perfect use case. Let's take your existing dbt models to production ML in just a few steps.

You already have a dbt model that computes driver metrics. All you need to do is add one tag to mark it for Feast:

tags=['feast'] # ← Just add this tag!

That's it. One tag. Your model doesn't change—it keeps working exactly as before for your analytics workloads.

You're probably already documenting your dbt models (and if you're not, you should be!). Feast uses this exact same documentation—no duplication needed:

This creates `target/manifest.json` with all your model metadata—the same artifact you're already generating.

Use the Feast CLI to discover your tagged models:

Now for the magic—automatically generate production-ready feature definitions from your dbt model:

In seconds, Feast generates a complete Python file with everything needed for production ML serving—all from your existing dbt model:

You just went from dbt model to production ML features without rewriting a single line of transformation logic. Your dbt model—with all its carefully crafted SQL, documentation, and testing—is now:

- **Serving features in milliseconds** for real-time predictions

- **Maintaining point-in-time correctness** to prevent data leakage during training

- **Syncing with your data warehouse** automatically as your dbt models update

- **Self-documenting** using the descriptions you already wrote

And here's the best part: when you update your dbt model (maybe you add a new column or refine your logic), just re-run `feast dbt import` and `feast apply`. Your production features stay in sync with your dbt source of truth.

Data teams using Feast with dbt are seeing real impact:

- **"We stopped rewriting features twice"**: Data engineers build once in dbt, ML teams use directly

- **50-70% faster ML deployment**: From dbt model to production features in minutes, not weeks

- **Single source of truth**: When dbt models update, ML features stay in sync

- **Analytics expertise becomes ML expertise**: Your dbt knowledge directly translates to ML feature engineering

- **Better collaboration**: No more "Can you rewrite this SQL in Python?" conversations

You've invested time and care into your dbt models. They're clean, documented, tested, and trusted by your organization. They shouldn't have to be rewritten to power machine learning—they should work as-is.

Feast's dbt integration makes that possible. Your dbt models become production ML features with:

- ✅ No rewriting or duplication

- ✅ No changes to your dbt workflow

- ✅ All your documentation preserved

- ✅ Real-time serving for predictions

- ✅ Point-in-time correctness for training

If you're a dbt user who's been asked to "make those models work for ML," this is your answer.

Ready to see your dbt models in production? Install Feast and try it out:

cd your_dbt_project

dbt compile

feast dbt import target/manifest.json -e your_entity_column -d bigquery

Your models are already great. Now make them do more.

Join us on [Slack](http://slack.feast.dev/) to share your dbt + Feast success stories, or check out the [full documentation](https://docs.feast.dev/how-to-guides/dbt-integration) to dive deeper.