⋈
01 / DUPLICATE JOINSOne purchase order.
Three times the quantity.
Test key cardinality and join grain. See which relation repeats a measure before changing the query.
GROUP BY po_id
HAVING COUNT(*) > 1BUILT FOR SMALL DATA TEAMS · EARLY MVPInvestigate incorrect data
Investigate incorrect data
with SQL you can review.
When a join multiplies quantities, a receipt goes missing, or two reports disagree — turn a hunch into a trail of approved queries and evidence.
✓ PostgreSQL & BigQuery✓ Local credentials✓ Explicit query approval
01 / INVESTIGATE
Schema.
SQL.
Evidence.
YOU KEEP THE CONTROLS.
A WORKBENCH, NOT ANOTHER GUESSThe question is only
The question is only
the beginning.
Explore your schema, review each SQL check, and see exactly which result supports a finding. One workspace. A visible evidence trail.
● ● ●QueryBridge / Local workbenchACTUAL APPLICATION · SYNTHETIC DATA
Captured from the working application, using the synthetic procurement fixture.Explore the workflow →
Built for the “why is this wrong?” moments.
One purchase order.
∅
02 / MISSING RECORDSA receipt exists.
Its purchase order doesn’t.
Propose unmatched-row checks across permitted tables. Separate missing source records from join-filter mistakes.
LEFT JOIN procurement p
WHERE p.po_id IS NULL≠
03 / MISMATCHED TOTALSTwo reports.
Two different answers.
Compare bounded aggregate checks and document what they establish. Full reconciliation is a planned workflow.
SUM(ordered_quantity)
vs. SUM(joined_quantity)LOCAL BY DESIGN. CONTEXT BY CONSENT.Your database credentials
Your database credentials
stay on your machine.
The React workbench connects to a backend on your own machine. The public website is separate and never handles database credentials.
Claude receives your selected SQL and permitted schema metadata. Query results require a second review: inspect the exact payload and choose aggregate-only or redacted results.
Read the data-flow details →Your database
Least-privilege permissions
Local workbench + backend
Scoped schema · Approved SQL · SQLite history
Claude API
Selected SQL + schema. Results only after approval.
WHAT WORKS TODAY
An honest starting point.
Implemented in the MVP
PostgreSQL and BigQuery adapters, a restricted read-only SQL subset, explicit execution approval, first-party local MCP tools, saved investigations, and a credential-free synthetic demo.
On the roadmap
SQL Server, Snowflake, Databricks, deeper reconciliation workflows, and broader SQL coverage. These connectors are not implemented today.
Validation stage
The demo is scripted and uses synthetic data. Live integrations require your own credentials and validation in your environment. There are no claimed customers, certifications, or partnerships.
A useful answer should come
with a query you can inspect.
QueryBridge is an independent product in development for small data engineering and BI teams. We’re looking for practitioners who want to help shape a clear, careful investigation workflow.
Founder contact: [email protected]Start a conversation ↗LET’S INVESTIGATE TOGETHERBring a question.
Bring a question.
Help shape the workbench.
Early access is for small teams exploring SQL investigation workflows. Tell us about a join, a missing record, or a total you don’t trust.
No production database access is needed to start a conversation.EARLY ACCESS BY EMAIL
Let’s talk about your data.
Send an email about your team and the investigation you want to try. This opens your email application; the website does not submit or store a registration form.
Request early access ↗[email protected] · Privacy notice