WrenAI

WrenAI is an engine for generative business intelligence (GenBI): AI agents use it to turn natural-language questions into SQL that is planned against a governed semantic layer, then to build and deploy dashboards from the results. It is designed to be driven by the agents you already use, such as Claude Code, Codex, or Cline, rather than through its own chat interface.

The semantic layer is expressed in the Modeling Definition Language (MDL): models, columns, relationships, views, cubes, and metrics that describe what the data means independently of any one warehouse. Alongside MDL sits an AI context layer of business definitions (instructions.md), approved example queries (queries.yml), and a local LanceDB memory index with hybrid retrieval. All of this lives as version-controlled files in your repository, so agents inherit reviewed business definitions instead of guessing from a schema.

Queries run through an Apache DataFusion based engine that connects to 22+ data sources, including BigQuery, Snowflake, PostgreSQL, ClickHouse, Amazon Redshift, Databricks, and DuckDB. Governed execution primitives (dry-plan validation, row limits, functions, and structured errors with hints) keep agent-generated SQL inside guardrails.

The workflow has three parts. Generate: the agent fetches MDL context, recalls similar past queries, writes governed SQL, and executes it with wren query. Deploy: the agent builds a browser-side dashboard app powered by wren-core-wasm and ships it to your own Vercel or Cloudflare Pages account. Know: the agent enriches the project with business context, either one question at a time or by reading a raw/ folder and proposing definitions. Workflow guides are served from the CLI itself with wren skills get , and a short discovery stub installed into the AI client teaches it to fetch them.

WrenAI is open core. The engine, MDL semantic layer, governed text-to-SQL, MCP server, CLI, and connectors are open source; row- and column-level security, user and group access control, the GenBI UI and dashboards, GenBI Apps, agentic mode, and AI-assisted context preparation are commercial features of Wren AI Cloud or self-hosted Enterprise Plus. The earlier Docker-based chat-first product is preserved on a legacy/v1 branch as Wren GenBI Classic.

Features

• Governed text-to-SQL: questions are planned against MDL with schema-aware retrieval and dry-plan validation before execution
• Semantic layer (MDL): models, columns, relationships, views, cubes, and metrics defined once and reused across every agent and query
• AI context layer: business definitions, approved example queries, and a local memory index stored as reviewable files in Git
• Dashboard deployment: agent-built browser-side apps on wren-core-wasm, deployable to Vercel or Cloudflare Pages with a shareable URL
• Data sources: 22+ connectors through an Apache DataFusion engine, with DuckDB bundled for local use
• Agent-driven CLI: wren skills get serves onboarding, enrich-context, and genbi workflow guides; wren ask --guided|--direct wraps a question for weaker or stronger agents
• Agent SDKs: wren-langchain for LangChain and LangGraph, wren-pydantic, and a reference Python integration for other stacks
• Execution guardrails: functions, dry-plan, row limits, and structured errors with hints
• Sample dataset: a bundled jaffle_shop dataset for trying the full flow without your own database