By SupabaseLearning Objectives
- Understand what database.build is and how it runs a real Postgres database in the browser
- Explain how its AI assistant turns plain-language requests into tables and queries
- Evaluate where an in-browser AI database sandbox is useful, and where it is not
What Is database.build?
database.build (formerly postgres.new) is a free, in-browser Postgres sandbox with a built-in AI assistant, made by Supabase. You can spin up an unlimited number of databases and build them by describing what you want in plain language — the AI scaffolds the tables, relationships, and sample data, and you can immediately query the result with regular SQL.
What makes it unusual is that there is no server. The entire database runs inside your browser tab.
💡Key Concept
PGlite — Postgres in WebAssembly. database.build is built on PGlite, a complete build of Postgres compiled to WebAssembly so it runs directly in the browser. Every query executes locally in the tab — there is no remote Postgres container and no network round-trip. That is why you can create unlimited throwaway databases instantly and work with them even offline.
How It Works
You start by telling the assistant what you want to build — "a database for a bookstore with authors, books, and orders," for example — and it generates the schema and inserts representative data. From there you can:
- Refine with conversation: ask for new columns, tables, or constraints in plain language
- Import your own data: drag and drop a CSV and database.build infers a table from its contents
- Query directly: run standard SQL against the database the AI built
- Visualize and export: inspect the schema and take your work with you
The original version paired PGlite with a hosted model; version 2 adds bring-your-own-LLM support, so you can point it at your own model provider rather than the default.
Pricing
database.build is free and open source. Because the database runs entirely in your browser via PGlite, there is no backend to pay for — the only external cost is whatever LLM provider powers the AI assistant when you bring your own.
Strengths
- Zero setup: A real Postgres database in a browser tab, with nothing to install or provision
- Natural-language modeling: Go from idea to working schema and sample data in one sentence
- Instant and disposable: Unlimited throwaway databases make it ideal for prototyping and learning
- CSV-to-table: Drag-and-drop import is a fast way to explore a spreadsheet as queryable data
- Open and portable: Open source, with bring-your-own-LLM in version 2
Limitations & Considerations
- Sandbox, not production: It is a prototyping and learning environment, not a hosted application database
- Browser-bound resources: Because it runs in the tab, very large datasets are constrained by the browser
- Agent autonomy: The assistant operates the database with broad permissions — fine for a sandbox, but a reminder of why production needs guardrails
- Not a Supabase project: Work here does not automatically become a deployed Supabase backend; it is a separate sandbox
Best Use Cases
| Task | Why database.build fits |
|---|---|
| Prototyping a schema | Describe the data model and get working tables in seconds |
| Exploring a CSV | Drop a spreadsheet in and query it as a real table |
| Learning SQL and Postgres | Practice against a real database with no setup or account |
| Sketching an app's data layer | Iterate on relationships before committing to a production backend |
| Teaching databases | Students get a real Postgres instance instantly, in any browser |
Getting Started
- Open database.build in any modern browser
- Describe the database you want in plain language and let the assistant scaffold it
- Drag in a CSV to turn existing data into a queryable table
- Run SQL directly against the generated database to explore or refine it
- In version 2, connect your own LLM provider if you prefer it over the default
🎯Tip
Use it as a thinking tool. database.build shines for quickly testing a data model or "what would this look like as tables?" questions. Once a design feels right, recreate it as migrations in a real Supabase project — see Supabase Vector and the Supabase MCP Server for the production path.
Key Takeaways
- database.build (formerly postgres.new) is Supabase's free, in-browser Postgres sandbox with an AI assistant
- It runs a real Postgres database in the browser via PGlite (Postgres compiled to WebAssembly) — no server, no setup
- Describe a database in plain language and the AI scaffolds tables; drag in a CSV to auto-generate schema; query with SQL
- It is a prototyping and learning tool, not a production backend — version 2 adds bring-your-own-LLM support
