Built for dbt teamsAutoDBTTM
Half your dbt code is dead.
We bury it safely.
For data leaders who know something's wrong, but can't quantify it yet.
Why this PR
95% SQL similarity, only filter scope differs
7 downstream consumers, all preserved
Saves $3,200/mo in warehouse spend
Proof attached
Equivalence report · 100% match
Blast radius · 17 readers verified
Backfill plan · idempotent
$47k
average monthly warehouse spend recovered
per team, in first 90 days
40%
of dbt models are never read after 90 days
sitting in your repo right now, costing money
minutes
from connecting your repo to your first PR
not days. AutoDBT scans and acts immediately
Three problems every data team has.
One tool that fixes all of them.
The bill nobody can explain
Spend went up 12% last month. Nobody knows which model caused it. AutoDBT traces every dollar to the model that spent it.
The change nobody wants to make
3 models doing the same thing. Nobody touches them — the last person who tried broke a CFO dashboard. AutoDBT proves the fix is safe before anyone merges.
The mess that comes back
You cleaned it up 18 months ago. Standards drifted. It's back. AutoDBT blocks violations at PR time, before they reach main.
AutoDBT opens PRs. Your team merges them.
Nothing is automated end to end. Every change is proposed, proven, and reviewed before it ships. Your team stays in control.
AutoDBT scans
Reads your repo and connects to your compute engine. Finds duplicates, dead code, and drift. Ranks everything by cost.
Opens a PR with proof
For each fix, it opens a GitHub PR with an equivalence report, blast radius map, and backfill plan attached. If anything looks off, the PR is blocked automatically.
Your engineer reviews and merges
Your team reads the proof, runs your existing CI, and merges when they are confident. Nothing ships without a human approving it.
Nothing automatic
AutoDBT never merges on its own. It opens a PR, attaches the proof, and waits. Your team decides what ships and when.
Validation runs on all steps
Equivalence checks, blast radius scans, and layer rule enforcement run at every stage. If anything fails, the PR is blocked before your team sees it.
Controlled refactor
Every change is scoped, reversible, and tied to a PR. 30-day quarantine before any deletion. One command to roll back anything, anytime.
See the debt in your repo.
Duplicates. Drift. Dead code. Mapped, ranked, and priced across your full model graph.
Catch near-duplicates
95% the same. One different filter. That's why your numbers don't tie.
Spot stale logic
One model updates. Siblings don't. We flag every stale copy.
Clear the graveyard
Zero reads in 90 days. Unused macros. See what they cost.
fct_orders_v2.sql
SELECT order_id,
customer_id, revenue_usd
FROM stg_orders
WHERE status = 'complete'
AND region = 'US'
fct_orders_enriched.sql
SELECT order_id,
customer_id, revenue_usd
FROM stg_orders
WHERE status IN ('complete',
'shipped', 'delivered')
| Cluster | Models | Similarity | $ overlap |
|---|---|---|---|
| fct_orders | 3 | 95% | $3,200 |
| dim_users | 4 | 88% | $1,840 |
| fct_sessions | 2 | 82% | $720 |
Duplicates. Drift. Dead code. Mapped, ranked, and priced across your full model graph.
Cluster analysis
Finds models sharing 80–99% of their SQL. Ranked by downstream impact and cost overlap.
Logic drift scan
One model updates. Siblings don't. We catch the drift before it hits the dashboard.
Dead code map
90 days zero reads. We list every model, macro, and source. Sorted by what it costs you.
Cost attribution
Warehouse spend, model by model. The expensive ones are usually the ones nobody owns.
Enforce it as you grow.
Standards enforced at PR time. Backlog ranked by dollars. Every change reversible.
Block at the PR
No BI reading from stg_. Enforced on every PR.
Top of the queue first
Ranked by dollars saved. Always know what's next.
Undo anything
30-day quarantine. One-command rollback. No regrets.
Layer violation detected
Looker explore 'Order Analytics' reads from stg_orders. Move to fct_orders.
Duplicate check
No new duplicates introduced in this PR.
Dead code scan
All new models have at least one downstream consumer.
1 # raw → staging → intermediate → marts → exposures
2 stg_* → may read raw_*
3 int_* → may read stg_*, int_*
4 fct_*, dim_* → may read int_*, stg_*
5 exposures → may read fct_*, dim_*
Standards enforced at PR time. Backlog ranked by dollars. Every change reversible.
Layer Enforcement
Five layers. Hard rules. Cross-layer imports get blocked before merge.
Cost-Ranked Backlog
Sorted by dollars saved. Always know what's next, what's safe, and what's worth it.
30-Day Quarantine
Soft-delete first. Anyone shouts in 30 days, restore in one click. No-one shouts? It's gone.
One-Command Rollback
Every change reversible. zingle rollback <pr>. Done in 60 seconds.
Book a demo
See what's hiding in your repo.
Point us at your dbt project. We'll show you the rest.