Start here
The whole product on one small dataset, with every Deep Dive: relationships, model, govern, quality, query, test data, publish, operate.
Open live demoProvisa / Live demos
Live demos
Each demo opens in your browser with its own data already loaded, the example queries to run, and the tour. No account, nothing to install.
App backends
Moving a Hasura app
An outfitter's storefront app, served by Hasura v2 from PostgreSQL, imported into Provisa in one step. Its tables, names, relationships, roles, row rules and write permissions work at once, and its queries are reshaped once. Then it does what Hasura could not, joining the sales team's Salesforce accounts to the orders under the same rules.
Open live demo
Product reviews
An electronics shop's catalog in MongoDB, its customers' reviews in Elasticsearch and who bought what in Neo4j, joined on the product. Which products do reviews say the battery dies on, and what else did their buyers buy?
Open live demo
One model, app and BI
A retailer's order-history screen and its BI dashboard read one model of PostgreSQL, MongoDB and an internal REST service. The app asks one nested GraphQL query as the signed-in customer; Apache Superset charts revenue and late shipments over the PostgreSQL wire protocol as the analyst.
Open live demoBI across sources
Cloud policy review
Is every disk attached to a virtual machine encrypted with a customer-managed key? The cloud inventory across Azure, AWS and Google Cloud, the ServiceNow incidents filed against it and the security policy itself, read from its web page by code, reviewed daily by a scheduled trigger.
Open live demo
Instant data lake
A city's open-data drop (311 complaints in monthly CSV files, building permits in Parquet, the budget in Excel and inspections in JSON) read in place as one governed catalog, with no load step. Which neighborhoods have the most complaints near active construction?
Open live demo
What is actually in this data?
The customer and order tables an outdoor retailer received when it bought a smaller one, profiled by Provisa's built-in Data Profiler: nulls, distinct values, ranges, formats, plausible types and duplicates, landed as governed tables.
Open live demo
Live order stream
A coffee shop's orders arrive live on a Kafka topic and are joined to its catalog in Postgres. Which regions placed the most orders in the last five minutes? A view refreshed by every batch of orders that lands answers it in SQL, GraphQL and REST, and a subscription watches it move.
Open live demo
Web shop logs
A web shop's nginx access log and its database's statement log, read as tables and joined to the shop's own data. You see the slowest pages and the SQL behind them, the requests that failed in the database, the customers who were served an error, and the products viewed but never bought.
Open live demoData science access
Predict customer churn
Which customers are about to cancel? A subscription business's customers and their usage, read into pandas in the browser as the analyst role, email addresses masked, and a scikit-learn model trained on them in the Notebook tab, from a notebook the demo ships.
Open live demo
Churn risk in production
Which customers are about to cancel, scored in production? The churn model ships as a Python package whose function is a table: joined to the customers, it ranks every one, emails masked for the analyst and only its own region's rows for the EU team, with a Node twin beside it.
Open live demo
Shakespeare's sonnets
Shakespeare's 154 sonnets in Postgres, embedded into Qdrant, searched by meaning and grown into a graph that shows the themes running through them and the lines where Shakespeare echoes himself.
Open live demo
World Cup, one graph
Wikipedia's finals and confederation tables and Kaggle's match and goal datasets disagree on country names. A short Python function matches every name with an LLM, and the four public sources become one governed graph, with West Germany's titles counted as Germany's.
Open live demoGive an agent your data
Give an agent your data
Can an AI agent be given a retailer's order data and held to what its user may see? In the MCP explorer, the agent an outside client would get, a West support agent's assistant reads only the West's customers with emails masked, the analyst's reads every region and no email, and each answer opens to the governed SQL that ran.
Open live demoData provisioning
Data room for a sale
A retailer selling itself gives each bidder's clean team governed access to its own order system and CRM until the deal date, with no exports. Bidders get aggregates, profiles, quality scores and masked samples, and every statement is audited. No customer identity leaves the room.
Open live demo
Data vault
A data vault declared over two live systems from Northwind, an order system and a CRM. Hubs, links and satellites are views the engine builds and refreshes as new rows arrive, personal data is masked in the satellites, and one point-in-time query answers what was known about a customer, from which source, and when.
Open live demo
Provisioning datasets to downstream teams
A coffee roaster's warehouse builds a dataset from its order, price and stock feeds as they land, and provisions it to finance and data science. Each team is told on Kafka when it is ready and pulls it over Arrow Flight as of that moment, seeing its own governed view, with no copies.
Open live demo
Data you can trust
A chain of shops' daily sales feed, landed into bronze and refined into silver and gold by materialized views that check every batch. Landing a batch is all it takes: a clean batch reaches gold, and a bad one is shown failed and never reaches silver.
Open live demoNew or changing sources
Weather and crop insurance losses
A crop insurer's portfolio of Corn Belt farms, in its own database, joined by county and crop year to the USDA's crop insurance losses by cause, NOAA's daily weather and NASS's county yields from government data. Does weather explain the losses in our portfolio?
Open live demo
Group exposure through Wikidata
A Prague bank's corporate loan book, in its own database, joined by each borrower's owner's Legal Entity Identifier to Wikidata's chains of parent organizations over SPARQL. Which corporate groups is the bank over its 25% large-exposure limit to? The bank, its borrowers and its loans are fictional; the groups that own them are real.
Open live demo
Property exposure and hazard risk
A property insurer's book of policies, in its own database, joined by county to FEMA's National Risk Index, the US Forest Service's wildfire risk and NOAA's storm damage from government data. Where is our property exposure concentrated against hazard risk?
Open live demo
Opioid settlement funds and overdose deaths
A state opioid settlement fund's payments to Ohio's counties, in its own database, joined by county to CDC's overdose deaths, the Census Bureau's population and federal spending from government data. Which counties have the most overdose deaths for the treatment the fund pays for?
Open live demo
Drug shortage and recall risk
A hospital pharmacy's formulary and stock, in its own database, joined by NDC and manufacturer to FDA's drug shortages and recalls and to Medicaid's prescriptions from government data. Which of our products face shortage or recall risk?
Open live demo
Exploited vulnerabilities in our code
A software company's repositories and the Dependabot alerts on their packages, in its own database, joined by CVE to the NVD and CISA's catalog of Known Exploited Vulnerabilities from government data. Which vulnerabilities attackers exploit sit in our code, and how far past CISA's due date are they?
Open live demoCompliance and audit
Insider review
A bank's outbound mail matched against its control room's watch list of deals in discussion, by name, ticker and code name and, through a local embedding model, by meaning, and checked for customer data. The result is one ranked review queue, and the reviewer never sees the watch list itself.
Open live demo
Reproduce a report
A bank's exposure report, reproduced for its regulator from a live trading book. It shows where a figure came from, column by column and refresh by refresh, and the report as it was filed, untouched by a correction recorded since. Every read of it is in the audit log.
Open live demo
One retail group, two regions
A retail group with an EU and a US subsidiary, run as one business in two regions under one model: can it do that without breaking GDPR? EU customer data stays in the EU, group reporting reads EU customers pseudonymised, and access and erasure requests, the insider review and the audit are each one query.
Open live demo
Test data
A shop's production database in three environments: prod, the real rows; pre-prod, production's rows with every sensitive column faked for every role; and dev, generated from production's profile with nothing real. The same queries, views and joins answer in all three.
Open live demo