Blog

Translates free-form natural language questions into executable SQL queries against connected databases using a semantic layer context engine. The system maintains a semantic model (either from dbt definitions or manual configuration) that provides table relationships, column meanings, and business logic, which the LLM uses to ground query generation and prevent hallucination. Queries execute in-place against source databases (Databricks, etc.) rather than copying data, enabling real-time analysis on current state.

Supports extended conversational workflows where users iteratively refine questions, ask follow-up questions, and build complex analyses across multiple turns. The system maintains conversation context and can decompose multi-step analytical tasks (e.g., 'show me sales by region, then drill into the top region, then compare to last year') into sequential SQL queries. Distinct from ad-hoc mode which optimizes for single-question speed; interactive mode trades latency for analytical depth.

Offers open-source deployment option enabling self-hosted installation and operation of Wren AI, providing data sovereignty and avoiding vendor lock-in. The system can be deployed on-premises or in private cloud environments, with source code available for customization and audit. This contrasts with cloud-only SaaS deployments and enables organizations with strict data residency requirements to use Wren AI.

Provides a semantic context engine designed to support AI agents and autonomous systems, enabling agents to understand data relationships, business logic, and query semantics. The context engine maintains semantic metadata (from dbt or manual definitions) and provides it to agents for grounding natural language understanding and query generation. This enables agents to reason about data and make autonomous decisions based on accurate information.

Embeds Wren AI's natural language query engine directly into Slack, allowing users to ask data questions and receive results without leaving the chat interface. Queries are executed against connected databases and results (likely visualizations or formatted tables) are posted back to Slack channels or DMs. This reduces context-switching friction for teams that use Slack as their primary communication hub.

Automatically ingests dbt project metadata (models, columns, descriptions, relationships, tests) as semantic context for query generation, eliminating the need to manually define a separate semantic layer. The system parses dbt's manifest.json and uses dbt model definitions, column documentation, and relationship definitions to ground natural language queries in actual data structure and business logic. This approach leverages existing dbt governance and documentation investments.

Executes natural language queries directly against Databricks lakehouse environments with native integration, including support for Databricks-specific features like Unity Catalog, Delta Lake optimizations, and Databricks SQL compute. Queries are translated to Databricks SQL dialect and executed using Databricks' query engine, enabling real-time analysis on lakehouse data without data movement.

Automatically generates visualizations (charts, tables, or other visual formats) from query results, presenting data in a human-readable format rather than raw SQL result sets. The system infers appropriate visualization types based on result schema and data characteristics (e.g., time series data → line chart, categorical aggregations → bar chart). Visualizations are rendered in the UI, Slack, or other output channels.

Automatically resolves complex multi-table joins based on semantic metadata (dbt relationships or manual definitions) without requiring users to explicitly specify join logic. When a natural language question references multiple tables, the system uses semantic relationship definitions to determine join paths and conditions, executing the necessary joins transparently. This enables users to ask questions across related tables without understanding the underlying schema structure.

Optimizes for single-question query latency by skipping multi-turn conversation overhead and executing queries with minimal context management. Ad-hoc mode is designed for quick, one-off questions where speed is prioritized over analytical depth. This contrasts with interactive mode which maintains conversation state and supports complex multi-step workflows at the cost of higher latency.

Provides specialized support for human resources analytics use cases, including employee data, compensation, headcount, and organizational structure queries. The system handles HR-specific data relationships (employee-department-manager hierarchies, compensation bands, tenure calculations) transparently, enabling HR teams to ask complex questions about workforce composition and trends without SQL expertise.

Provides specialized support for supply chain analytics use cases, including inventory tracking, supplier performance, lead time analysis, and anomaly detection. The system handles supply chain-specific data relationships and calculations, enabling supply chain teams to identify issues (e.g., leak detection in supply chain data) and optimize operations without SQL expertise.