Wren AI

Open-source core generative BI agent that converts natural language to SQL using a semantic layer (MDL). The commercial and open-source versions run on the same context engine, supporting cloud, VPC and air-gapped deployment.

AI Developer Tools#Open Source#Text-to-SQL#BI

Updated: 2026-10-26

What is Wren AI?

Wren AI adopts an open-core approach: the "Context Engine" responsible for understanding your questions is open-sourced on GitHub (with approximately 17.6K stars). The company emphasizes that all commercial solutions run on the same engine—whether on the cloud or self-hosted, it's identical, with "No black box, no lock-in." It aims to solve the problem of "LLMs don't know your data" (the official wording: LLMs don't know your data. The open core does.). It uses a semantic layer (MDL) to map business terms to actual tables and fields, ensuring natural language queries target accurate data structures. The official team also provides a comparison guide covering 12 platforms and 22 selection factors for users to reference during evaluation.

Who is it for?

Data teams that want to enable business colleagues to query data independently without being locked into commercial BI platforms; enterprises with mandatory requirements that data must not leave the intranet (supports air-gapped deployment); and organizations that want to first validate the open-source version and then smoothly upgrade to a paid solution.

Key features

Text-to-SQL Agent

Ask questions in natural language, generate executable SQL and return results; the core TEXT-TO-SQL, MCP and CLI are all open-source components.

Semantic Layer (MDL)

The mapping between business terms and database structures is the key to accuracy; the MDL resides directly in your Git repository, enabling versioning, auditing, and collaboration. You can even build the semantic model offline first before integrating it into the engine.

Generative BI Applications

On the basis of open-source engines, the commercial version superimposes Agent mode, context preparation and governance capabilities. The product is available in three forms: Wren AI OSS, Wren AI Cloud and Enterprise Plus, with open and transparent boundaries.

Multi-deployment modes

Supports cloud, in-VPC, and air-gapped environments. The official advocates "Start where your constraints are".

Readable and Modifiable

The core logic is open-source, allowing teams to read the code and run it locally; the official team also specifically lists "questions skeptics should ask" and discloses the open-source/commercial boundaries.

How to use Wren AI

First, clarify the definitions of core metrics and high-frequency entities in MDL, then open it up for business parties to ask questions.
The quality at the semantic level directly determines the experience and deserves priority investment.
before selection, confirm each item against the official "Questions Skeptics Should Ask" (such as what to do after stopping payment, whether it is possible to migrate from open source to paid).
verify the accuracy rate in a small-scale pilot before promoting it to the entire organization.
For scenarios sensitive to result accuracy, you can first conduct a pilot with a read-only account and manually verify for a week, then gradually open it up.
this also applies to small organizations where the data team is small but want to open data capabilities in a low-threshold way.

Things to know

The ceiling for accuracy lies in the semantic layer rather than the model; if the MDL definition is incomplete, the generated SQL will frequently contain errors.

The open-source version is just the engine; the full Agent mode, context preparation, and governance capabilities are available in the commercial version. Pay attention to the boundaries between open-source and commercial when evaluating.

Maintaining the semantic layer requires continuous investment; business term mappings must be updated promptly when business terminology changes; for databases with complex table structures or confusing naming conventions, the workload of organizing the MDL in the early stage should not be underestimated.

Self-hosting requires you to prepare models, computing power and operation and maintenance resources by yourself; the trial and payment terms of the cloud version are subject to the latest instructions on the official website, while the open source version is completely free with no usage restrictions.

When write operations are involved in SQL generation, be sure to confirm the read-only policy and permission configuration to prevent accidental modification of data.