AI assistants and search decision
AI2sql
A technical user can reasonably build a usable NL→SQL assistant covering single-user generation, explanation, read-only execution and a simple UI in about a week, but reproducing AI2sql’s agent-facing MCP gateway, desktop app, audited metering, team features and enterprise polish is more work and not fully covered by this small replacement.
Visit website↗$9/mo
$108/yr
Read off the official pricing page.
$100one-off36 h to build
$50/mo3 h/mo upkeep
On cash alone, building overtakes the subscription at 7 seats.
Open-source builds that already do this
Every project below is open source and already does this job today. Fork one, self-host it, or take the parts you need - the build prompt further down assumes an empty file, and this is the shortcut past that. Licences differ; check the one on each card before you ship. All AI2sql alternatives, with the arithmetic →
What a replacement has to do
- Accept a DB connection, introspect schema, accept plain-English question, generate dialect-aware SQL via an LLM, enforce read-only checks, execute query and return rows.
What it still won’t have
- Governed MCP endpoint for AI agents (agent-facing MCP integration)
- Desktop app (macOS/Windows) for purely local execution
- Built-in audited gateway with scoped API keys and per-query metering
- Priority/team features (shared query library, RBAC, execution priority) and packaged enterprise offerings
- Polish, training/tuning of their advanced model and product support
What remains hard
- Product polish and ongoing maintenance
First-year cost
Keep paying
Paying is—cheaper in year one.
On cash alone, building overtakes the subscription at 7 seats.
Money you would actually spend
Time you would spend
—
What you would spend
What we assumed
The verdict above measures whether you could build it. This one is only about money.
Runnable build prompt
Build a minimal schema-aware NL→SQL assistant using Node.js/Express, React, Postgres (for metadata caching), and OpenAI (or other LLM) API. Core features in scope: (1) DB connector to accept read-only credentials and introspect schema (tables, columns, FKs) and store metadata in Postgres; (2) a REST API endpoint that accepts plain-English questions, constructs prompts referencing the cached schema, calls an LLM to generate dialect-aware SQL and a plain-language explanation; (3) a safety classifier that blocks non-read-only statements (DDL/INSERT/UPDATE/DELETE/multi-statement) before execution; (4) execution layer that runs read-only queries via node-postgres (or mysql driver) and returns rows, timing and row count; (5) simple React UI to enter questions, preview generated SQL and explanation, and display results. Out of scope: multi-database desktop app, MCP gateway for agents, multi-tenant team/RBAC, advanced query optimization using execution plans. Require error handling for DB/connectivity/LLM failures, input validation, logging, and automated tests for schema introspection, prompt→SQL flow, safety blocker, and query execution.
How we checked
How the score was reached
- Partly verdict base52
- An open-source build was found+5
- 5 cited sources+3
- Price verified on pricing page+3
- Evidence score63
The base comes from the verdict. Everything under it is a check that either happened or did not, and each one is a fact frozen in this record rather than a judgement made at render time - so the same evidence always produces the same number.
How scoring works →Cited sources · 5
Every page the run actually retrieved.
- official productAI2SQL — product
- official pricingAI2SQL — pricing
- official docsAI2SQL — MCP for Agents
- open sourcecloudera/hue
- open sourcechartbrew/chartbrew
Integrity checks
What held up, and what did not.
