TL;DR: Bruin is the best text-to-SQL tool for teams that need answers they can trust, because it writes the SQL on data it loads, transforms, and checks itself, and attaches column-level lineage to every answer. The other text-to-SQL tools worth knowing in 2026 are WrenAI, Snowflake Cortex Analyst, Databricks Genie, ChatGPT's Data agent, Claude, Querio, Vanna AI, and Defog. They split into four groups: open-source engines and models you wire up yourself (WrenAI, Vanna AI, Defog's SQLCoder), warehouse-native NL-to-SQL built into the platform (Snowflake Cortex Analyst, Databricks Genie), general AI assistants that now write SQL against a connected warehouse (ChatGPT's Data agent, Claude), and AI analytics platforms where text-to-SQL is one step inside an answer (Bruin, Querio). Generating SQL was never the hard part. Trusting the result is, which is why the tools that pair text-to-SQL with governance, quality checks, and lineage are the ones business teams actually rely on.
Text-to-SQL went from research demo to table stakes in 2026. Almost any model can turn "show revenue by region last quarter" into a query. The real question is whether the answer is correct, consistent, and safe to act on, and that depends on far more than the SQL generation itself. In September and October 2026 the general AI assistants joined in: OpenAI's Data agent for ChatGPT Work and Anthropic's Claude Dashboards both write SQL against Snowflake, BigQuery, or Databricks from a plain-language question. This guide groups the tools by how they actually work and where text-to-SQL fits.
We build Bruin, so you know where we stand. We also say plainly where another tool is the better pick: for an engineer embedding NL-to-SQL in an app, or a team that lives entirely in one warehouse, it often is. For the broader picture see the best AI data analyst tools and AI data analyst vs ChatGPT, Claude, and coding agents.
Key takeaways
- Writing the SQL is the easy part. Almost any model can draft a query; the hard part is knowing it used the right table, the right definition, and data that passed its checks.
- There are 4 kinds of text-to-SQL tool. Open-source engines, warehouse-native features, general AI assistants with warehouse connectors, and AI analytics platforms, and each one leaves a different amount of work to you.
- 2 of the biggest AI assistants now write SQL on your warehouse. ChatGPT's Data agent (September 2026) and Claude (October 2026) query Snowflake, BigQuery, and Databricks under your own permissions, but neither checks the data underneath.
- Pick by who asks the question. Engineers embedding NL-to-SQL want a library they control; business teams want an answer with the SQL and lineage attached.
How we compared
- What happens after the SQL is written: does the tool check the data and show where the answer came from?
- Open source or managed, and who runs it.
- Which data it can reach: one warehouse, several, or SaaS sources too.
- Who it is built for: engineers embedding it in an app, or business teams asking questions.
Facts about each tool come from its public documentation, checked in November 2026. Facts about Bruin come from our docs and pricing page.
What is a text-to-SQL tool, exactly?
A text-to-SQL tool converts a natural-language question into a SQL query against your database or warehouse. The naive version just prompts a model with your schema. The useful version also:
- Grounds the query in your real schema, metric definitions, and past queries, so it is accurate.
- Connects to live company data rather than a sample.
- Validates results against quality checks, so it does not answer from broken data.
- Returns consistent answers, because "revenue" maps to the same definition every time.
- Shows the generated SQL and lineage, so the answer is trustworthy.
The four kinds of text-to-SQL tool
- Open-source engines and models. WrenAI's engine, Vanna AI, and Defog's SQLCoder models give engineers building blocks to add NL-to-SQL to their own apps. Maximum control, and you build the trust layer.
- Warehouse-native. Snowflake Cortex Analyst and Databricks Genie put NL-to-SQL inside the warehouse you already run.
- General AI assistants with warehouse connectors. ChatGPT's Data agent and Claude write and run SQL against a connected warehouse inside the assistant your company already uses. Fast to start, and only as reliable as the warehouse behind them.
- AI analytics platforms. Bruin and Querio treat text-to-SQL as one step toward an answer people can check, not the end product.
The 9 text-to-SQL tools at a glance
| Tool | Type | Open source | Connects to live data | Beyond SQL (governed answer) | Best for |
|---|---|---|---|---|---|
| Bruin | AI data analyst + platform | Core CLI (Apache 2.0) | Yes | Yes | Trustworthy answers and dashboards, not just SQL |
| WrenAI | Open-core GenBI engine | Engine yes (Apache 2.0) | Yes | Partial: semantic layer | Open-source NL-to-SQL with a semantic layer |
| Snowflake Cortex Analyst | Warehouse-native | No | Yes (Snowflake) | Partial: semantic views | Teams whose data lives in Snowflake |
| Databricks Genie | Warehouse-native | No | Yes (Databricks) | Partial: curated Genie Agents | Teams whose data lives in Databricks |
| ChatGPT Data agent | General AI assistant | No | Yes (Snowflake, BigQuery, Databricks, Redshift, ClickHouse, MongoDB) | Partial: reads dbt or Snowflake Horizon definitions | ChatGPT Work customers with a governed warehouse |
| Claude | General AI assistant | No | Yes (Snowflake, BigQuery, Databricks, Redshift, ClickHouse) | No: SQL per chart | Quick questions in Claude on a modelled warehouse |
| Querio | Git-native AI analytics | No | Yes | Partial | Data teams that want versioned analytics with a Slack bot |
| Vanna AI | Python library (repo archived in 2026) | Yes (MIT) | Yes (you wire it) | No | Engineers who will maintain the code themselves |
| Defog | AI data analyst + open SQLCoder models | Models are open-weight | Yes | Partial | Self-hosting an open text-to-SQL model |
Best text-to-SQL tool by need:
- Trustworthy answers grounded in governed data, quality checks, and lineage, in Slack, Teams, or the browser: Bruin
- Open-source NL-to-SQL engine with a semantic layer to embed or self-host: WrenAI
- A small MIT-licensed library you will maintain yourself: Vanna AI
- All-in Snowflake teams: Snowflake Cortex Analyst
- All-in Databricks teams: Databricks Genie
- Companies standardized on ChatGPT Work with a governed warehouse: ChatGPT Data agent
- Companies standardized on Claude that want quick charts: Claude
The 9 tools, one by one
1. Bruin
What it is: an AI data analyst on an end-to-end platform. Text-to-SQL is one step: Bruin generates the query, runs it against your governed data, checks quality, and returns a trustworthy answer in plain English, with the SQL and column-level lineage attached.
Where it wins: the SQL is grounded in your real schema, metric definitions, and lineage, so answers are consistent for everyone. You ask from Slack, Microsoft Teams, Google Chat, WhatsApp, or the browser, and Bruin can also build the dashboard. Because the same platform loads and checks the data, it can tell you when last night's load failed before it answers.
Where it falls short: it is a platform, not a library. If you only want an NL-to-SQL component to embed in your own app, an open-source library is lighter, and Bruin asks you to set up the pipeline in code.
Best for: business and data teams that want answers they can trust, with the pipeline included. See the Bruin AI data analyst.
2. WrenAI
What it is: an open-core GenBI engine from Canner. The engine (the MDL semantic layer, text-to-SQL, an MCP server, a CLI, and connectors) is Apache 2.0; the GenBI interface, dashboards, and row-level security come with Wren AI Cloud or the self-hosted Enterprise Plus edition.
Where it wins: the semantic layer gives the model your business definitions, and the engine is open source and actively developed, so you can embed or self-host it.
Where it falls short: the full interface is commercial, and you still model the semantic layer and keep the warehouse loaded and tested yourself.
Best for: teams that want an open-source NL-to-SQL engine with a semantic layer.
3. Snowflake Cortex Analyst
What it is: Snowflake's natural-language-to-SQL feature over Snowflake data, grounded in a semantic view (YAML semantic models are the older option), with a REST API for building your own apps. Snowflake now points new projects toward Cortex Agents, which can use it.
Where it wins: it runs inside Snowflake's own governance and role-based access control, so there is no new system to secure.
Where it falls short: it only sees Snowflake, and it does not load your sources, build dashboards, or check the data across the rest of your stack.
Best for: teams whose data lives in Snowflake.
4. Databricks Genie
What it is: Databricks' natural-language interface. A data team sets up a Genie Agent (formerly called a Genie space) with tables, instructions, and example SQL, and business users ask questions in Genie One.
Where it wins: answers respect each user's own Unity Catalog permissions, inside the platform your data already lives in.
Where it falls short: it only sees Databricks, and answer quality depends on how well the data team curates each Genie Agent.
Best for: teams whose data lives in Databricks.
5. ChatGPT Data agent
What it is: OpenAI's Data agent for ChatGPT Work, launched on 10 September 2026. You type @Data, it writes and runs SQL against Snowflake, BigQuery, Databricks, Redshift, ClickHouse, or MongoDB under your existing permissions, and it can build dashboards.
Where it wins: it reads metric definitions you already wrote in dbt or Snowflake Horizon, inherits your warehouse permissions, and has no separate price for ChatGPT Work customers.
Where it falls short: it does not load, model, or check data, and OpenAI tells users to verify the source, period, filters, and definition behind an answer. See ChatGPT vs Bruin.
Best for: ChatGPT Work customers with a governed warehouse.
6. Claude
What it is: Anthropic's assistant, which since the Dashboards beta of 8 October 2026 writes a SQL query per chart against Snowflake, BigQuery, Databricks, Redshift, or ClickHouse.
Where it wins: every chart shows its query and when it last refreshed, you can ask Claude to explain any number, and queries run through each viewer's own connector.
Where it falls short: each chart writes its own SQL, so there is no shared metric definition unless the warehouse encodes one, and it cannot tell whether the tables passed their checks. More in Can Claude Dashboards replace your BI tool?
Best for: companies that run on Claude and want quick charts on a modelled warehouse.
7. Querio
What it is: a warehouse-connected AI analytics platform that describes itself as Git-native, built for data teams and their AI agents, with a web app, notebooks, dashboards, a Slack bot, and MCP, CLI, and API access.
Where it wins: analysis lives in Git with a review history, and agents can reach it through MCP or the CLI.
Where it falls short: it sits on a pipeline and warehouse you run elsewhere. See Querio vs Bruin.
Best for: data teams that want versioned, warehouse-connected analytics.
8. Vanna AI
What it is: an MIT-licensed Python framework for text-to-SQL that grounds an LLM agent in your schema and past queries, with a hosted Vanna Cloud. Its GitHub repository was archived in 2026, and the last release was v2.0.2 in February 2026.
Where it wins: it is small, permissively licensed, and easy to embed in your own application.
Where it falls short: with the repository archived, you maintain the code yourself or move to Vanna Cloud, and accuracy, governance, and the answer experience are yours to build.
Best for: engineers who want a simple library and are happy to own it.
9. Defog
What it is: defog.ai sells an AI data analyst for business users, and the company published the open-weight SQLCoder text-to-SQL models, last updated in 2024.
Where it wins: SQLCoder gives developers an open model they can run on their own hardware.
Where it falls short: the open models have not been updated since 2024, and the hosted analyst answers from a warehouse you keep loaded and modelled elsewhere.
Best for: developers who want to self-host an open text-to-SQL model.
A note on Seek AI, which earlier versions of this guide listed: in June 2025 IBM took on Seek AI's team and licensed its technology for watsonx AI Labs, so it is no longer a separate product to evaluate.
Which text-to-SQL tool should you pick?
| Your situation | Pick | Why |
|---|---|---|
| Business teams need trustworthy answers in Slack, Teams, or the browser | Bruin | SQL on data it loads and checks, with lineage attached |
| You are an engineer embedding NL-to-SQL in your own app | WrenAI | An Apache 2.0 engine with a semantic layer, MCP server, and CLI |
| You want a small MIT library and will maintain it yourself | Vanna AI | Simple to embed; the repository was archived in 2026, so you own updates |
| Your data lives entirely in Snowflake | Snowflake Cortex Analyst | Answers inside Snowflake's own governance |
| Your data lives entirely in Databricks | Databricks Genie | Curated Genie Agents that respect Unity Catalog permissions |
| Your company runs on ChatGPT Work with a governed warehouse | ChatGPT Data agent | Reads dbt or Snowflake Horizon definitions you already wrote |
| Your company runs on Claude and wants quick charts | Claude | The fastest path from a question to a chart in a tool people already use |
When Bruin is not the right pick
If you are building your own product and want NL-to-SQL as a component, WrenAI's open engine gives you more control with less to adopt. If every table you care about already lives in Snowflake or Databricks and is well modelled, Cortex Analyst or Genie is the shorter path. Bruin earns its place when the data behind the question is not yet loaded, modelled, and checked in one place, or when business teams need the answer, the SQL, and the lineage together.
FAQ
What is the best text-to-SQL tool in 2026?
Bruin is the best fit for business teams that need trustworthy, consistent answers, because it pairs text-to-SQL with governed data, quality checks, and lineage. For embedding NL-to-SQL in your own app, WrenAI's Apache 2.0 engine is the leading open-source option. For all-in Snowflake or Databricks teams, Cortex Analyst and Genie work in-platform.
Is there an open-source text-to-SQL tool?
Yes. WrenAI's engine is Apache 2.0, Vanna AI is MIT-licensed (its repository was archived in 2026), and Defog's SQLCoder models are open-weight. They give engineers control to build NL-to-SQL into their own applications. Bruin's core CLI is also open source (Apache 2.0), with text-to-SQL as part of the managed AI analyst layer on top.
Can ChatGPT or Claude write SQL against my data warehouse?
Yes, as of autumn 2026. ChatGPT's Data agent (ChatGPT Work, September 2026) and Claude (Dashboards beta, October 2026) both write and run SQL against Snowflake, BigQuery, Databricks, and other warehouses under your own permissions. Neither loads, models, or checks the data, so the answer is only as reliable as the warehouse. Bruin generates the SQL on data it loads, transforms, and checks itself, with column-level lineage attached.
Why is generating SQL not enough?
Because the hard part of data analysis is trusting the result, not writing the query. A model can produce SQL that runs but uses the wrong table, a stale definition, or data that failed validation. Tools that ground the SQL in your real schema and metric definitions, check quality, and expose lineage, such as Bruin, are what make the answer reliable enough to act on.
Which text-to-SQL tools connect to live company data?
Most serious options connect to live data: WrenAI, Vanna AI and Defog's models (once wired up), Snowflake Cortex Analyst, Databricks Genie, ChatGPT's Data agent, Claude, Querio, and Bruin. The warehouse-native ones are scoped to their own platform; Bruin connects across warehouses and SaaS sources and keeps the answer governed.
What is the difference between a text-to-SQL tool and an AI data analyst?
A text-to-SQL tool produces a query. An AI data analyst produces a trustworthy answer: it generates the SQL, runs it on governed data, checks quality, returns the result in plain English with lineage, and can build dashboards or take action. Bruin is an AI data analyst, so text-to-SQL is one step rather than the whole product.
See trustworthy text-to-SQL in action
If you want answers grounded in governed data, not just generated SQL, see how Bruin works, check pricing, or read AI data analyst vs ChatGPT, Claude, and coding agents.