SQL-executed, not LLM-inferred
A question is converted into SQL that traverses the discovered relationship graph. The join was mapped and scored before the question was ever asked, not invented in response to it.
Platform Data Foundation
One question. Every system. An answer you can check.
Ask something that spans a database, a spreadsheet, and a document archive that nobody designed to talk to each other, and get an answer with the SQL that ran, the tables it touched, and a confidence score attached, not a fluent, confident, wrong paragraph you have to take on faith.
cross-source query executed
Question
Which customers have open invoices but no recent orders?
Generated SQL
SELECT c.name
FROM customers c
JOIN invoices i ON i.customer_id = c.id
LEFT JOIN orders o ON o.customer_id = c.id
WHERE i.status = 'open' AND o.id IS NULL Answer
source attribution confidence score row count
Overview
Most AI-on-your-data tools will attempt a cross-source answer. Few will show their work. The distinction that matters isn't whether the answer sounds right, it's whether you can trace it back to an executed query against a warehouse whose relationships were already discovered and scored, versus a language model inferring a join on the fly and hoping it holds.
Source attribution on every answer
Which tables, which sources, which rows. An answer without a visible source is treated as a defect, not a feature.
Confidence score, visible
The same scoring that underlies the relationship graph surfaces at the point of answer, so a low-confidence join produces a visibly lower-confidence result rather than the same fluent tone as a high-confidence one.
Row count, always shown
An answer built on three matching rows looks different from one built on thirty thousand. Both are shown as what they are.
This is what Relationship Discovery and Ingestion & ETL were building toward, and it inherits the same access boundaries as Governed Warehouse: a role that can't see a table can't get an answer derived from it, no matter how the question is phrased.
Insights
Use cases & field notes from the data trenches.
Customer playbooks, technical deep-dives and field notes from deployments into messy enterprise estates.
Blog
baselineproof of valueoutcome
A 30-day proof of value, not a chatbot demo
A demo proves a question can be asked. A proof of value establishes whether the right sources, rules and evidence produced an answer that changed a decision.
Read the article Technical
triggercandidatesevidencereview
Investigate the reason behind the alert
Without investigation, an organisation risks automating the distribution of unexplained alerts. The result is more noise, not better decisions.
Read the article Blog
signal context evidence next step
An alert is not an insight
More alerts do not create more value. Without context and prioritisation, they simply create another queue of work.
Read the articleConnectors
Reads from the systems you already have.
Databases, warehouses, files and the apps you run: connect what you already have, don't migrate. Every source lands in one governed graph, with 200+ on the roadmap.
Don't see your connector? Tell us your stack. We'll scope it.
Reach out →