BlazeSQL

BlazeSQL is an AI data analyst for SQL databases and warehouses. Users ask questions in natural language, and the assistant generates queries, retrieves results and presents tables or charts. It can turn useful results into dashboards, maintain knowledge about business definitions, and run scheduled agent reports. Technical and non-technical interfaces let a team choose how much SQL detail users see.

The official website and help center describe desktop, web and embedded workflows. BlazeSQL is suitable for companies whose data already lives in SQL systems and whose analysts or business users need to explore it. A useful deployment combines its conversational interface with reviewed metric definitions, appropriately restricted database access and a process for checking important answers.

Connecting the right database and tables

The connection guide lists systems including PostgreSQL, MySQL, SQL Server, Snowflake, BigQuery and Databricks. Users can connect a supported database directly or provide table and column names for SQL generation that they run manually. Direct connections enable the assistant to retrieve live results.

Connection requirements depend on the database and application surface. The desktop app can connect from a device on the database's private network. The web app needs an appropriate reachable connection and supports an SSH-tunnel option. Confirm the supported authentication method and network route with the database administrator before configuring access.

Select the tables relevant to the analysis. A small, understandable set of customer, subscription and payment tables is easier to reason about than an unrelated collection spanning every system. Use a database account with suitable read access and avoid giving a conversational analysis tool broader permissions merely to resolve a connection problem.

Schema changes need maintenance. The database management guide distinguishes refreshing table and column metadata from updating connection credentials. After a migration, refresh the selected structure and review questions that depend on renamed or removed fields.

Technical and non-technical conversation modes

Technical mode displays SQL and explains query logic. It may ask for details about joins or filters. This suits analysts and developers who can inspect the generated statement and provide precise corrections.

Non-technical mode runs SQL in the background and emphasizes the business interpretation. Simplified mode also hides elements such as database selection, table structure and model settings. These interfaces can make the product easier for business users to navigate, but hiding SQL does not remove the need for sound definitions and reviewed setup.

Describe the question with its measurement, period and grouping. A request for customer growth should specify whether it means newly created accounts, new paying customers or active subscriptions. Include whether the date range is based on account creation, payment or another event. Those choices can produce substantially different valid queries.

For a consequential answer, ask a technical reviewer to examine the underlying logic. A readable explanation can make a mistaken assumption sound reasonable. Check the query and a small set of source records when the result will influence staffing, forecasting or a published report.

Knowledge Notes and reviewed training questions

The Knowledge Page guide describes Knowledge Notes that store definitions, rules and context. Notes can be scoped to a database, schema, table or column. They help explain details that table names and relationships cannot establish, such as how a particular business defines a qualifying transaction.

Training Questions provide a review workflow: ask a representative question, inspect the answer, give feedback and review suggested notes before saving them. Query Review helps technical users and administrators inspect team conversations and move problematic requests into that training workflow.

A useful note expresses a reusable rule. If trial accounts are excluded from a customer metric, document that definition and its relevant field. A note containing only the expected total for one month does not explain how to calculate the metric next month. Keep unrelated rules in separate notes with appropriate scope.

Review changes to shared knowledge as a team asset. An accepted correction can influence later answers for other users. When the business changes its definition, update the relevant note and recheck representative questions. Knowledge maintenance is part of keeping analytics consistent, even after the initial connection works.

From query results to saved dashboards

The dashboard builder lets users add graphs or tables from chat and arrange them with a drag-and-drop editor. This can turn an exploratory question into a recurring view without separately rebuilding every result in another dashboard tool.

Saved insights can also be changed through a conversation. The saved-query guide describes opening an existing graph or query, asking the chatbot to modify it, and either saving a new item or overwriting the original. That distinction matters when colleagues rely on a shared dashboard.

Give each chart a clear measure and time basis. A daily trend should distinguish transaction date from settlement date, and a revenue chart should state how refunds are handled. Confirm that the displayed aggregation matches the intended business question before saving it for others.

Preserve an established insight when exploring a different view. Save a separate result for a new region or alternate definition until the team accepts the change. Otherwise an exploratory request can silently replace the meaning of a familiar dashboard item.

Scheduled Agent Reports

Agent Reports analyze selected topics on a schedule. Users write a prompt, optionally attach saved queries or dashboards, run a test and then activate delivery. The report view shows formatted findings; the analysis view shows the queries and supporting work behind them.

Scheduling includes a time and timezone, with available frequency and delivery options depending on access. Reports can go to email or configured Slack and Microsoft Teams destinations. A significance filter can suppress delivery when the agent finds no meaningful update, while the suppressed report remains available in the app.

Define what the report should monitor and what action a reader can take. A prompt to watch every important business change is harder to review than one asking for weekly sign-up trends by acquisition source. Attach accepted queries where they provide a stable measurement, and examine the test run before enabling recurring delivery.

Treat an explanation of a change as a hypothesis to assess. A rise in cancellations near a product release does not by itself prove the release caused it. Inspect segment sizes, missing data and other events. Assign an owner to investigate significant findings and to distinguish analysis failures from delivery failures.

Desktop data locality and web processing

The privacy policy distinguishes data flows. With desktop offline mode enabled, query results stay on the device while schema metadata supports SQL generation. Disabling that mode sends results for deeper AI analysis. The web app processes and stores results on the service for features such as dashboards and collaboration.

Offline mode therefore does not mean an entirely offline AI model. Metadata, chat content and account information still have service-processing implications. Optional extraction of categorical values also adds information beyond table and column names. Confirm the actual settings and workflow rather than inferring privacy behavior from the word desktop.

The security overview describes the vendor's encryption and AI-provider retention arrangements. The privacy policy also explains credential handling and authentication exceptions. Review these with the organization's requirements, and avoid assuming that infrastructure certifications automatically establish a particular certification for BlazeSQL itself.

Choose the surface around the needed feature and acceptable data flow. Keeping results local can limit the assistant's ability to inspect them for deeper analysis. Collaboration may involve different processing. Those are practical tradeoffs to verify during setup, not settings to change solely because one option appears more convenient.

Workspace roles, groups and row restrictions

Workspace roles distinguish administrators, users and dashboard viewers. Administrators manage the workspace and have broad data access. Simplified business-user interfaces and viewer roles serve different needs from the technical staff maintaining the knowledge base and connections.

Access groups can control databases, schemas, tables and columns. Group permissions combine as a union, so membership in another group can grant additional access. Administrators are exempt from these restrictions, and the documentation distinguishes chat restrictions from dashboard-viewer behavior. Review both group membership and the sharing surface.

Row-level controls match a selected table column against a member's configured filter identifiers. Only configured tables are filtered. The guide states that a member with no filter identifiers does not receive a row-based filter, and newly added tables need their own configuration.

The same guide explicitly excludes integrations such as Slack, Teams, ChatGPT and Claude, as well as Agent Reports, from these row-level controls. Do not assume a chat restriction carries into those paths. Test each intended delivery and embedded workflow with representative user accounts before exposing tenant-specific information.

Embedded analytics and external integrations

BlazeSQL offers white-label embedded analytics for products that want a branded data-chat interface. The embedding guide describes appearance configuration, generating a per-user authenticated session URL and embedding the interface in an application. This requires appropriate subscription access.

The user-management API supports creating and updating users, database access and session links. Authentication keys belong in the application's backend. User identity and tenant access should come from the product's authenticated state rather than from an arbitrary email address supplied by a browser request.

Dynamic Integrations use a connected service's API credentials so the agent can interact with it. The documentation describes making these integrations available to the workspace. Confirm the credential's scope and begin with a limited retrieval task before granting operations that alter external data.

An embedded chat or API connection needs its own review. Verify the correct user, database and row scope; check session expiration and removal of access; and test what happens when credentials are revoked. Branding the interface does not replace those application responsibilities.

A concrete workflow for subscription analytics

Suppose an operations team wants to understand weekly subscription changes. A technical owner first identifies the authoritative account, subscription and event tables, creates a suitably restricted connection and checks which records represent tests or canceled transactions.

1. Define new paying customers, cancellations and reactivations in Knowledge Notes, including the date field and exclusions for each measure.
2. Use technical mode to ask representative questions and inspect joins, filters and returned rows. Correct the definitions before broader use.
3. Review suggested notes from Training Questions, then invite a small group with the appropriate role and database access.
4. Save accepted weekly tables or charts to a dashboard. Keep alternate segment views separate while they are being evaluated.
5. Create an Agent Report using those accepted sources, test its analysis and confirm that its delivery audience can receive the included data.
6. Monitor real questions and refresh metadata after schema changes. Revisit shared definitions when the subscription business rules change.

Reconcile a sample of cancellations and reactivations with known accounts. One customer can have several subscriptions, and a cancellation event may precede the end of service. A query can run successfully while counting the wrong entity. That check makes the report useful for the team's actual decision.

Pricing, suitability and practical limitations

BlazeSQL is a paid product with a trial, and its website explicitly says there is no permanently free tier. Individual pricing and team pricing distinguish access levels. Advanced analysis, APIs, embedding, recurring reports and enterprise administration have plan-dependent availability. Confirm current capacity and terms for the intended workflow.

It fits data and operations teams that have SQL sources and a technical owner able to validate setup. Non-technical users benefit most from clear shared definitions and an accepted initial set of questions. Organizations without SQL data may be better served by a spreadsheet or file-analysis workflow.

Practical limitations include ambiguous metrics, outdated schema metadata, incorrect joins, changing source quality and differences between application surfaces. Generated explanations and dashboards can look convincing despite a faulty query. Vendor claims about accuracy or speed should be evaluated on the organization's own reviewed questions.

Frequently asked questions

Must every user know SQL?

No. Non-technical mode emphasizes business explanations. A technical owner should still configure connections, validate definitions and review important queries.

Can it generate SQL without a live connection?

The connection guide describes providing table and column names for generation, with queries run manually. Live retrieval requires an appropriate connection.

Does desktop offline mode keep all processing on the device?

No. It keeps query results local in the documented mode, while metadata and service interactions still support AI generation. Check the chosen settings and data flow.

Do row restrictions apply to scheduled reports?

The current row-control guide explicitly excludes Agent Reports and listed integrations. Review access separately for those workflows.