Text-to-SQL for teams that need more than a notebook

What it does

Dataherald is a monorepo that turns plain-English questions into SQL queries against your relational databases. It exposes the core engine through an API and layers on enterprise features—authentication, organizations, and user management—so you can embed it in a SaaS product or let business users interrogate a data warehouse directly. The stack also includes a Next.js admin console for configuration and observability, plus a Slack bot for channel-based queries.

The interesting bit

Most open-source text-to-SQL projects stop at the model layer; Dataherald treats the unglamorous parts—auth, multi-tenancy, and a GUI—as first-class citizens. It is largely integration plumbing, but it is plumbing that saves you from rebuilding the same enterprise scaffolding around an LLM call.

Key highlights

• Core engine can run standalone if you just need the NL-to-SQL API without users or authentication
• Enterprise service adds authentication, organizations, and users on top of the engine, along with other business logic
• Admin console provides a GUI for configuration and observability, though it requires both engine and enterprise services to function
• Slack bot lets channel members query databases conversationally, also requiring the full backend stack
• Apache 2.0 licensed and packaged as Docker services

Caveats

• The README is light on technical details about the actual NL-to-SQL implementation—supported models, fine-tuning strategy, and accuracy benchmarks are not mentioned
• The admin console and Slack bot have hard dependencies on both the engine and enterprise services; there is no standalone mode for the front-end or chat integrations

Verdict

Worth a look if you need to ship a natural-language database interface to non-technical users and would rather not build auth, observability, and chat integrations from scratch. Skip it if you are just experimenting with prompt engineering or need a lightweight, single-process library.