Natural-language database questionsPublic field guide
DataChat: natural-language questions to inspectable SQL
DataChat is a conversational workbench for databases. A person starts with a business question, while the application uses schema context to produce a query that can be inspected. The aim is to bring the question, the query, and the answer into one understandable workflow.
Start with a practical evaluationThe useful starting point
What DataChat gives your team
An engineer and a business user can discuss the same visible query and result, making hidden assumptions easier to identify before a number is used.
Data teams, business-software developers, and organizations exploring an AI interface to their databases.
- 01
Ask a question
Describe the business result and clarify ambiguous terms.
- 02
Inspect the query
Review generated SQL against the schema and permitted access.
- 03
Check the answer
Compare returned values with an independent reference.
Business language needs a precise interpretation
A request such as 'show our best customers' has several possible meanings. It might refer to revenue, margin, retention, or recent activity. A natural-language interface is useful when the underlying definition becomes visible instead of being silently guessed.
Schema context helps the model work with the database's tables and fields. The generated query then becomes an artifact the team can review. A correct-looking query is still only one stage: its returned values need to answer the intended question.
Query generation and execution are different responsibilities
The demonstrated DataChat path connects Claude to SQL Server. Its implementation includes query checks and execution controls such as cancellation and timeouts. Those mechanisms make a useful source evaluation possible; they do not replace database permissions.
The application also contains other provider and database adapters. Each combination needs a live test with its own credentials, schema, and deployment environment. The demonstrated external-model route is not an offline or air-gapped deployment claim.
Current public evidence
Working data conversation stack
The application has demonstrated its Claude-to-SQL Server path and includes schema-aware generation, dangerous-query blocking, resource checks, streaming, cancellation, timeouts, priorities, cache, history, audit metrics, multiple model providers, and adapters for major databases. Every adapter still earns its own live proof.
Explore the project and its evidenceTry one bounded question
Ask one question with a known answer
A suggested evaluation for your team.
- Use a small synthetic SQL Server dataset and a least-privilege evaluation account.
- Define one business measure, including dates, refunds, and missing values.
- Write the expected answer independently before asking DataChat.
- Review the generated query, compare values, and repeat the request.
- Test one ambiguous request, a denied operation, and cancellation.
A visible connection between business meaning, SQL, and the returned result. Keep incorrect answers and operational failures alongside successful cases.
Before you go further
Common questions about DataChat
Which database path has been demonstrated?
The public evidence identifies Claude to SQL Server. Other adapters are implementation material that still needs its own live evaluation; their presence does not prove every combination works.
Can nontechnical people use the results?
The conversational approach starts with ordinary language, but an organization's metric definitions and permissions still matter. An initial trial should pair a business user with someone who can independently verify the query and answer.
What should we test before connecting company data?
Review authorization, provider data handling, query execution controls, and logging. Use synthetic data first. The linked evaluation guide provides a more detailed repeatable protocol.
From an idea to your first evaluation
Build on what you understand.
Start with DataChat's public evidence. If it fits your team's problem, compare source memberships or tell us what you would like to evaluate.
These guides are free to read. Private source releases follow the membership license. Compare plans and team seats.