live wire
▸nl2sql.ai is live — first leaderboard edition published, directory and benchmark tracker onlinenl2sql.ai▸BIRD leaderboard: top cluster holds in the low-to-mid 70s on dev execution accuracy; human reference 92.96bird-bench.github.io▸Spider 2.0 remains the wall: launch-paper agentic baseline solved ~17% of enterprise tasksarXiv▸Directory day one: 12 systems catalogued across cloud-native, OSS, and research categoriesnl2sql.ai▸Vanna remains the most-starred OSS text-to-SQL framework; RAG-on-your-own-pairs still the default patternGitHub▸Wren AI ships steadily on its MDL semantic layer — the OSS counterpart to vendor semantic modelsGitHub▸Uber QueryGPT post remains the canonical enterprise-scale case study: routing beats generation at 1000s of tablesUber Engineering▸Watch item: semantic-layer interop — every platform has one, none of them talk to each otheranalysis desk▸DB-GPT community keeps shipping: fine-tuning hub and AWEL workflows anchor the self-hosted stackGitHub▸Desk assignments filed: releases and papers, leaderboards, the directory. Cadence: continuousnl2sql.ai▸nl2sql.ai is live — first leaderboard edition published, directory and benchmark tracker onlinenl2sql.ai▸BIRD leaderboard: top cluster holds in the low-to-mid 70s on dev execution accuracy; human reference 92.96bird-bench.github.io▸Spider 2.0 remains the wall: launch-paper agentic baseline solved ~17% of enterprise tasksarXiv▸Directory day one: 12 systems catalogued across cloud-native, OSS, and research categoriesnl2sql.ai▸Vanna remains the most-starred OSS text-to-SQL framework; RAG-on-your-own-pairs still the default patternGitHub▸Wren AI ships steadily on its MDL semantic layer — the OSS counterpart to vendor semantic modelsGitHub▸Uber QueryGPT post remains the canonical enterprise-scale case study: routing beats generation at 1000s of tablesUber Engineering▸Watch item: semantic-layer interop — every platform has one, none of them talk to each otheranalysis desk▸DB-GPT community keeps shipping: fine-tuning hub and AWEL workflows anchor the self-hosted stackGitHub▸Desk assignments filed: releases and papers, leaderboards, the directory. Cadence: continuousnl2sql.ai
nl2sql.ai
guideTOOLS DESK

Field guide: the open-source NL2SQL stack

Vanna, Wren AI, DB-GPT, the Dataherald engine, SQLCoder — what each is actually for, and how to choose.

By The Tools Desk· Aug 10, 2026

The open-source text-to-SQL ecosystem sorts into four rough roles. Most production stacks combine two or three.

The framework: Vanna

Vanna (MIT) is the default starting point: a Python RAG framework you train on your own DDL, documentation, and known-good SQL pairs. Model-agnostic, vector-store-agnostic, database-agnostic. Its insight — that your organization's past queries are the best grounding data you own — became the standard OSS pattern. Choose it when you want a library, not a product.

The semantic-layer product: Wren AI

Wren AI (AGPL) is the open-source answer to Cortex Analyst: it builds an explicit semantic modeling layer (MDL) over your database, then serves a full GenBI experience — chat, SQL, charts, summaries — through a web UI. Choose it when the consumers are business users and you want grounding you can version-control.

The platform: DB-GPT

DB-GPT (MIT, eosphoros community) is the kitchen-sink option: local model support, a text-to-SQL fine-tuning hub, multi-agent orchestration, and the AWEL workflow language. Popular where data cannot leave the building. Choose it when you're building data applications, not just answering questions — and budget time for its surface area.

The reference engine: Dataherald

The Dataherald engine (Apache-2.0) survives its company as one of the cleanest reference architectures in the space: API-first NL2SQL with context stores, golden-SQL feedback loops, and evaluation hooks. Choose it to study or fork a production-shaped service design.

The models: SQLCoder

Defog's SQLCoder line proved that SQL-specialized open-weight models compete with far larger generalists on this task. Pair one with Vanna or your own pipeline when latency, cost, or data residency rule out frontier APIs.

The honest caveats

Every one of these inherits the field's open problems: semantic drift as schemas evolve, evaluation on your data (public leaderboard rank transfers poorly), and the wrong-but-plausible failure mode that no framework fully solves. The projects that take those seriously in their docs tend to take them seriously in their code.

The full directory — licenses, links, and coverage history — lives on the tools page, updated as the ecosystem moves.

sources

Filed by The Tools Desk. Corrections: desk@nl2sql.ai · Our standards →
more from the desk