Pinning packages in 2026
The problem
A packages.yml copied from a blog post installs whatever “latest” meant that week, or a git URL with no revision, or a package that is no longer maintained. Official dbt documentation treats packages as full dbt projects: after dbt deps, their models, macros, and tests become part of yours.
A working example
This site’s companion project used the official Hub form and a pinned version:
packages:
- package: dbt-labs/dbt_utils
version: "1.4.1"
dbt deps on November 18, 2026 installed dbt-labs/dbt_utils 1.4.1 from the Hub (https://hub.getdbt.com/dbt-labs/dbt_utils). The official Hub API reported 1.4.1 as latest that day. dbt run then used dbt_utils.generate_surrogate_key successfully.
Official docs distinguish packages.yml from dependencies.yml:
- Package dependencies pull someone else’s dbt project in like a library.
- Project dependencies are for Mesh / cross-project
ref. - On dbt v1, Jinja in package specs (for example a git token) belongs in
packages.yml. Official docs say dbt v1 does not render Jinja independencies.yml.
Official docs also warn if you point at a git repo with no pin:
The git package "https://github.com/dbt-labs/dbt-utils.git" is not pinned.
This can introduce breaking changes into your project without warning!
They mention warn-unpinned: false and immediately say that is not recommended.
A Hub package with a version range is still a pin; official docs commonly show ranges like ">=1.0.0", "<2.0.0" when you want patches but not a new major.
What to be cautious about in 2026
This is not a complete market survey. It is a short list of things that were checked while researching this site:
dbt_utils— still the standard utility package; pin a Hub version and read the changelog when you bump it.dbt-expectations(calogica/dbt-expectations) — the GitHub README, fetched November 18, 2026, states the package is no longer actively supported. It had 1,235 stars that day. Stars are not a maintenance guarantee.- Elementary — an actively used dbt-native observability package/product (2,435 GitHub stars that day). Evaluate it as a product, not only as a handful of tests. Not installed here.
- Fivetran source packages and similar vendor packages — useful if you load those extractors; they will add models to your run. Disable what you do not need.
Always read the package’s own README for the version you pin, and official Hub / docs for install syntax.
The gotcha
Official troubleshooting: an old dbt_utils that still uses config-version: 1 raises a runtime error under current dbt. The fix they document is to update the version in packages.yml to a current Hub release.
Installed package models will run unless you disable them. A “I only wanted one macro” install can still materialize tables in your warehouse.
When not to use this
- Do not add a package to solve a one-line
coalesce. Write the SQL. - Do not leave git URLs unpinned in a shared project.
- Do not install an unmaintained package because an old tutorial named it.
dbt deps installed dbt_utils 1.4.1, matching the Hub API latest field on November 18, 2026. Pinning, unpinned-git warnings, and packages-vs-dependencies notes are from official dbt documentation. The dbt-expectations support notice and star counts are from GitHub, same day.