In LangChain SQL agent
You want full control of the agent loop, prompts, tools, and retrieval.
Comparison
nlqdb vs LangChain SQL agent
Pick the LangChain SQL agent if you want to build and own the agent loop — its prompts, tools, retries, and deployment — over a database you already run. Pick nlqdb if you want NL→SQL working today as a hosted pipeline you embed, with the SQL shown, writes diff-previewed, and a Postgres provisioned for you.
The same goal, two ways.
> monthly active users for the last 6 months
In your HTML
<nlq-data goal="monthly active users for the last 6 months"></nlq-data>A from-scratch SQL agent needs a wired DB, model, prompt, and guardrails before this works; nlqdb answers it from one English goal over a Postgres it provisioned, SQL shown.
What's different
Four dimensions that decide it
| Dimension | nlqdb | LangChain SQL agent | Note |
|---|---|---|---|
| Provisions + owns the database (from English) | The LangChain SQL agent connects to a database you already stood up; it provisions nothing. nlqdb materialises a Postgres from the first English goal. | ||
| NL→SQL without building or tuning the agent yourself | LangChain gives you the toolkit and a ReAct loop; you assemble, prompt, and tune the agent. nlqdb ships the pipeline — nothing to wire. | ||
| Compiled SQL shown with every answer | You can log the agent's intermediate SQL, but surfacing it to a user is your wiring. nlqdb returns the compiled SQL with every answer by default. | ||
| Fail-closed read-only SQL validation | SQLDatabaseToolkit's query tool runs whatever SQL the model emits; guardrails are yours to add. nlqdb validates against an allowlist and fails closed. |
Show 5 more rows
| Dimension | nlqdb | LangChain SQL agent | Note |
|---|---|---|---|
| Destructive-op diff preview before apply | LangGraph can pause for human approval before a tool call, but it computes no before/after diff of a write. nlqdb diff-previews writes and DDL. | ||
| Plan cache (repeat questions skip the LLM) | Caching identical questions is yours to build in LangChain. nlqdb caches the compiled plan so a repeated question returns without another model call. | ||
| HTML embed element + anonymous try | nlqdb ships `<nlq-data>` and anonymous mode (a first answer before sign-in); a LangChain agent is code you deploy and host yourself. | ||
| Self-host / open source | LangChain is open source and self-hostable; nlqdb's platform is hosted-only during pre-beta (SDKs / CLI / elements open at GA). | ||
| 100s of framework integrations (models, stores, loaders) | LangChain's reach across models, vector stores, and loaders is its moat; nlqdb is Postgres-first NL→SQL, not a general agent framework. |
shipped · partial · not shipped
When to choose nlqdb
- You want NL→SQL working today without building a prompt, validator, retry, and eval stack.
- You want the compiled SQL shown with every answer so you can audit the grain.
- You want a Postgres provisioned from English — no database to stand up first.
- You want destructive writes and migrations diff-previewed before they apply.
When to choose LangChain SQL agent
- You want full control of the agent loop, prompts, tools, and retrieval.
- You're already in the LangChain / LangGraph ecosystem with other agents.
- You need it free, open source, and self-hosted against a database you run.
- You want to wire your own few-shot examples, tools, and reasoning graph.
Questions buyers ask
- Is nlqdb a replacement for the LangChain SQL agent?
- Only for the NL→SQL job. The LangChain SQL agent is a framework you assemble — toolkit, model, prompt, retries, deployment — over a database you already run. nlqdb is the hosted pipeline that does NL→SQL for you, provisions the Postgres, shows the SQL, and diff-previews writes. One is build-it-yourself; the other is embed-and-go.
- Can I use the LangChain SQL agent and nlqdb together?
- Yes. A LangChain agent can call nlqdb as one of its tools — letting nlqdb own the database and the NL→SQL step while LangChain orchestrates the wider reasoning, retrieval, and other tools. You get nlqdb's validated, SQL-shown answers without hand-building the SQL sub-agent.
- Does the LangChain SQL agent provision a database for me?
- No. SQLDatabaseToolkit connects to a database you've already created and configured; it lists tables, reads columns, and runs SQL against it. nlqdb materialises a Postgres from your first English goal, so there's nothing to stand up before the first question.
- What do I have to build myself with the LangChain SQL agent?
- The model choice, the prompt, the tool loop (or LangGraph graph), SQL guardrails, retries, caching, deployment, and evaluation (LangSmith, separate). That flexibility is the point. nlqdb ships those as a managed pipeline, so you trade tunability for not maintaining the stack.
- Does nlqdb show the SQL the way a LangChain SQL agent logs it?
- nlqdb returns the compiled SQL with every answer by default, so a user or auditor can check the grain. A LangChain agent's intermediate SQL is in its trace, but surfacing it cleanly to an end user is wiring you write yourself.
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 →