Exmergo logoExmergo

Explore.Transform.Maintain.

The agent-native analytics engineering toolkit.

Point it at your warehouse and your dbt project. It learns the landscape, authors your transformations, and tells you exactly what to fix when the schema drifts.

npx skills add exmergo/dex

Codex, Gemini CLI, Cursor, and others, via AGENTS.md.

dex: thelook_ecommerce
ranked inventory
order_items181K rows
inventory_items490K rows
users100K rows
products29K rows
PIIusers.email flagged PII (email) · never surfaced
SELECT product_id, COUNT(*) AS units_sold, SUM(sale_price) AS revenue FROM order_items JOIN inventory_items ON …
needs confirmation · est. 49.3 MB scanned
allowed · 10 rows returned
Draft map cached to .dex/ · fully read-only

Connects to your warehouse

Snowflake logoSnowflake
Databricks logoDatabricks
BigQuery logoBigQuery
Redshift logoRedshift
DuckDB logoDuckDB
PostgreSQL logoPostgreSQL

Connects to your Semantic Layer

dbt logodbt

Data and analytics teams are stuck between a rock and a hard place.

Two meters run on every query an agent writes. Dex is made to hold both of them down.

The warehouse bills every scan

Snowflake and Databricks are some of the most expensive consumption-billed software on the planet. An agent left to explore freely explores on your credit card.

billed per byte scanned

The agent bills every token

Anthropic and OpenAI bill by the token, and nothing burns through a context window faster than data analytics. One unguarded schema dump can cost more than the query it was meant to plan.

billed per token generated
Open source · Apache 2.0

Dex is made for teams paying both meters.

  • Dex forces the agent to clear a cost guard before it runs an exploration query or a transformation.
  • Dex gives the agent a set of tooling that makes it hard to get burned when transforming data.

Both come from tight control scripts that the SKILL.md files point the agent at, on every /dex:explore, /dex:transform and /dex:maintain.

The guardrails are not a tax on quality.
76%
on ADE-benchwith Claude Sonnet 5
2.5x
cheaper to runthan Claude Fable 5
57.3%
top published Sonnet 5 resultat k=1 on Snowflake’s 103-task data-eng-bench
See the benchmarks

Analytics Engineering is ETM

ETL ends at Load. Dex is built for what comes after.

01Explore
/dex:explore

Make sense of the warehouse.

Stop running SELECT * LIMIT 10 by hand. Dex intelligently profiles columns, infers relationships, and builds a comprehensive draft map of your tables, cached locally. PII is automatically flagged, never surfaced.

inferred relationships

order_itemsusersproducts
primary key verified · joins probed
02Transform
/dex:transform

Author dbt and semantic models.

Ask for a metric, and Dex writes the reviewable diffs. It authors dependency-aware dbt SQL models, schema tests, and full MetricFlow YAML. Validate against dev targets with built-in cost guards before anything runs.

dbt

thelook_semantic.yml

+ semantic_models:

+ - name: order_items

+ metrics:

+ - name: revenue

+ - name: return_rate

validated · cost-guarded
03Maintain
/dex:maintain

Detect drift before it breaks.

Warehouses change. Business definitions drift. Run /dex:maintain against your .dex/ snapshot to instantly detect schema, grain, and semantic drift. Dex proposes the exact edits needed to keep your project correct.

3 findings

Schema drift

products.brand column no longer found. Closest matches: brand_name, manufacturer_name

Volume drift

stg_order_items: 7-day avg 866 rows/day → 43 rows in last load (-95%)

Grain drift

fct_order_items lost uniqueness on order_item_id: 1,148 duplicates (fan-out from a new coupon_redemptions join)

3 reconciling edits proposed, ready for review
Security

Your raw data never leaves the warehouse.

Dex is read-only against your data and never routes it through Exmergo. Security is architectural, not a setting.

Aggregates leave, rows do not.

Dex is read-only, and understanding is built from SQL aggregates: schema, statistics, and bounded summaries cached locally in .dex/. Raw rows and flagged PII are never copied out. Exactly what crosses is written down in the Dex privacy policy.

Your agent, your keys.

You run Dex through your own coding agent, on your own API key or subscription. Exmergo never sits between your data and a model.

Credentials never reach the LLM.

Dex connects with your warehouse credentials directly (for example, BigQuery Application Default Credentials). Those credentials, and your raw data, never enter the model's context.

Built for the production stack.

Every scan is estimated with a dry-run and confirmed before it spends, then recorded in a local ledger.

Single-player is free. The living organization is Enterprise.

Seamlessly bridge the gap between analytics engineering and business consumption.

Dex

Open · Free · Apache 2.0
  • Full ETM loop (explore, transform, maintain)
  • Manual, on-demand drift detection
View on GitHub

Exmergo Enterprise Platform

Everything in free, plus:

  • Continuous, scheduled drift detection & automated PRs
  • Live, governed author-to-serve loop uniting Dex and Exmergo Viz
  • Cross-team governance, PII policy, and access audit
Contact Sales

Ready to turn maintenance into an automated habit?

Install the open-source toolkit today.

npx skills add exmergo/dex

Codex, Gemini CLI, Cursor, and others, via AGENTS.md.