In Dataherald
You already run a data warehouse and want NL queries over it.
Comparison
nlqdb vs Dataherald
Pick Dataherald if you have a data warehouse and want an open-source engine that answers it in English, tuned with business context and golden-SQL pairs you curate. Pick nlqdb if you want the database itself — a Postgres provisioned from English, the compiled SQL shown, and writes diff-previewed, with nothing to host.
The same goal, two ways.
> revenue by product category last quarter
In your HTML
<nlq-data goal="revenue by product category last quarter"></nlq-data>A warehouse-reporting question Dataherald answers over a warehouse you run; nlqdb answers it over a Postgres it provisioned, SQL shown.
What's different
Four dimensions that decide it
| Dimension | nlqdb | Dataherald | Note |
|---|---|---|---|
| Provisions + owns the database (from English) | Dataherald connects to a warehouse you already run; it provisions nothing. nlqdb materialises a Postgres from the first English goal. | ||
| Natural-language → SQL | |||
| Compiled SQL shown with every answer | Both return the SQL — it's the output. nlqdb shows it inline with the answer; Dataherald returns it from the engine API. | ||
| Connects to an existing warehouse (Snowflake / BigQuery / …) | Bring-your-own-warehouse is Dataherald's whole shape; nlqdb provisions + owns a Postgres instead (ClickHouse is Phase 2). |
Show 6 more rows
| Dimension | nlqdb | Dataherald | Note |
|---|---|---|---|
| Golden-SQL / training-pair tuning to your schema | nlqdb's plan cache + schema fingerprint are automatic; Dataherald's golden-SQL + fine-tune loop is explicit and curated. | ||
| Auto-migration via NL ('add a column for tags') | |||
| Destructive-op diff preview before apply | Dataherald answers (reads); nlqdb diff-previews writes and DDL before they apply. | ||
| HTML embed element + anonymous try | |||
| MCP server (agent-callable) | Dataherald exposes a REST engine API and a LangChain integration; it ships no dedicated MCP server. | ||
| Open source / self-hostable |
shipped · partial · not shipped
When to choose nlqdb
- You want NL→SQL working today without standing up and training an engine first.
- You want a Postgres provisioned from English — no warehouse to connect first.
- You want destructive writes and migrations diff-previewed before they apply.
- You want one embed and an MCP server, not an engine to host.
When to choose Dataherald
- You already run a data warehouse and want NL queries over it.
- You want to add business context and golden-SQL training pairs yourself.
- Open source and self-hosting matter for compliance or data residency.
- You need Snowflake, BigQuery, Redshift, or Databricks dialects today.
Questions buyers ask
- Is nlqdb a replacement for Dataherald?
- Only for the NL→SQL job, and from the other end. Dataherald is an open-source engine you self-host (or call as a hosted API) over a warehouse you already run, tuned with business context and golden SQL you curate. nlqdb is the hosted pipeline that owns the Postgres, compiles NL→SQL with the SQL shown, and diff-previews writes — nothing to host or train.
- Can I point Dataherald at an nlqdb database?
- Not the supported shape. nlqdb's API speaks `/v1/ask` (English) and `/v1/run` (raw SQL with an allow-list); Dataherald expects a warehouse connection string and its own context store. The honest split is nlqdb for a new data feature, Dataherald over a warehouse you can't migrate.
- Do I have to curate golden SQL like in Dataherald?
- No. Dataherald lifts accuracy by curating golden-SQL training pairs and business context against your schema. nlqdb prompts directly from the live schema fingerprint plus a recent-tables hint, and its plan cache fills automatically — there's no training set to author or maintain.
- Dataherald connects to my warehouse — does nlqdb?
- Not yet. nlqdb provisions and owns a Postgres rather than federating an existing warehouse; ClickHouse is Phase 2 and bring-your-own-connection is roadmap. If you must query Snowflake, BigQuery, or Redshift in place today, Dataherald is the better fit.
- Which one is better for an AI agent?
- nlqdb ships an MCP server with `nlqdb_query`, `nlqdb_list_databases`, and `nlqdb_describe`, where `nlqdb_query` materialises Postgres on first reference — so an agent stands up and queries its own data with no separate create-DB step. Dataherald exposes a REST API and a LangChain integration, so an agent wraps that API itself.
Try nlqdb in 30 seconds
No sign-in. The anonymous database lasts 72 hours; adopt it with one click if you keep it.
Start with a goal →