// the find
fal-ai/dbt-fal
do more with dbt. dbt-fal helps you run Python alongside dbt, so you can send Slack alerts, detect anomalies and build machine learning models.
dbt-fal was a bridge that let you attach Python scripts to dbt models — run them before/after model execution, pull model data into pandas with ref(), and write ML predictions back to the warehouse. It also added a Python model adapter for databases that dbt doesn't natively support with Python (Redshift, Postgres). The team abandoned it in April 2024 to pivot to generative media at fal.ai.
The ref() syntax in fal scripts was genuinely nice — you could pull a dbt model into a pandas DataFrame in one line without any SQL connection boilerplate. The structured hooks (before/after model run) were well-thought-out and the Behave-based integration test suite covered a lot of real adapters (Snowflake, Redshift, BigQuery, DuckDB, Athena, Postgres). Environment isolation via the isolate library was a real feature, not an afterthought — you could pin different Python deps per model.
It's dead. The repo is archived, no security patches after April 2024, and the team explicitly says there are no alternatives. Adopting this today means you own the fork. The two-component architecture (CLI + adapter) was also confusing — they solve overlapping problems in different ways and the docs don't make the boundary obvious. The cloud execution path was tightly coupled to fal.ai's commercial serverless offering, so the 'scale out' story evaporates now that they've pivoted.