Catalog as code.
Built for AI.
Backlogs is the open-source semantic layer where analytics engineers define what matters to the company in a versioned catalog that compiles to SQL. Anyone on the team can query governed metrics or contribute new ones — in natural language, through their preferred AI assistant, as PR-ready commits.
The dashboard era is ending
Your team adopted AI for code, writing, and research. But to get a number from the data warehouse, they still need a dashboard, a SQL query, or a ticket to the data team.
Requests pile up
Every "can you pull this?" becomes a ticket. Your analytics team is a bottleneck, not a force multiplier.
Consumers can't self-serve
BI tools require training. SQL requires skill. The people who need data most are the ones who can't get it.
Metrics drift
"Revenue" means something different in every dashboard. Without a governed source of truth, trust erodes.
Define once. Query anywhere.
Backlogs connects your analytics team's definitions to the AI assistants your whole company already uses.
Define as code
Analytics engineers define metrics, dimensions, and marts in a versioned YAML catalog — with AI assistance. Reviewed in PRs like any other code change.
Compile to dbt
Backlogs validates your catalog and compiles it to tested dbt models. No new runtime — it works with the warehouse you already have.
Query in conversation
Anyone on the team asks their AI assistant a question. Backlogs generates governed, production-quality SQL from natural language — no manual query writing needed.
Natural language in,
governed SQL out
When a PM asks "what was revenue by region last quarter?", their AI assistant doesn't hallucinate an answer. It calls your Backlogs catalog, assembles a query from your governed metric definitions, and returns production-quality SQL with real results.
Every query traces back to the definitions your analytics team reviewed and approved. Same metric, same formula, every time — regardless of who's asking or which AI they're using.
- Works with Claude, ChatGPT, or any MCP-compatible assistant
- SQL is deterministic — same question, same query, every time
- Consumers never write SQL — and can't break anything
Consumer asks their AI assistant:
"What was revenue by region last quarter?"
Backlogs generates governed SQL:
SELECT region, SUM(revenue)
FROM mart_sales
WHERE order_date BETWEEN
'2026-01-01' AND '2026-03-31'
GROUP BY region
Results returned:
| region | revenue |
|---|---|
| North America | $2.4M |
| EMEA | $1.8M |
| APAC | $920K |
Two sides of the same platform
Analytics engineers build the catalog. Everyone else queries it. Proposals bridge the gap — consumers request new metrics, engineers review and ship them.
For analytics engineers
Define your metrics layer with AI assistance in your editor. Create sources, transforms, metrics, dimensions, and marts — validated and compiled to dbt automatically.
- ✓ YAML catalog, version-controlled
- ✓ Compiles to tested dbt models
- ✓ Warehouse introspection and SQL preview
- ✓ Review consumer proposals inline
For data consumers
Ask your AI assistant a question and get back real results grounded in governed metrics. Save queries, set alerts, and propose new metrics when you need something that doesn't exist yet.
- ✓ Natural language to governed SQL
- ✓ Saved queries and threshold alerts
- ✓ Propose metrics your team is missing
- ✓ Works with any MCP-compatible AI assistant
No new runtime. No lock-in.
Backlogs compiles to dbt and queries via DuckDB. Run it on your laptop or deploy to your own AWS account.
See it in 30 minutes
We'll walk through defining a metric catalog, compiling it, and querying it in conversation — on real data.