Comparison
nlqdb vs Wren AI
Pick Wren AI if you already run a warehouse (BigQuery, Snowflake, PostgreSQL, DuckDB) and want a semantic model — models, metrics, cubes, RLAC/CLAC — governing every English question an AI agent asks. Pick nlqdb if you want the database itself provisioned, schema evolved via English, and answers rendered inside your app from one HTML element.
Persona this comparison serves: P3 analyst. Wren AI's positioning: Open-source context layer for AI agents — Modeling Definition Language (MDL) semantic model plus row- and column-level access controls over your existing warehouse.
When to choose nlqdb
- You need a new database — nlqdb provisions Postgres on the first query.
- You want destructive operations and schema changes diff-previewed before they apply, even when triggered by English.
- You want one HTML element rendering the answer in-product, not a hosted analytics surface.
- You need an agent calling MCP `create_database` — Wren AI's skills query an existing warehouse.
When to choose Wren AI
- You already run a production warehouse (BigQuery, Snowflake, PostgreSQL, DuckDB, …) you can't migrate.
- Your team needs a defined semantic model — metrics, cubes, relationships — governing every NL question.
- Row- or column-level access controls must apply before any AI agent reads a row.
- SOC 2 Type II is required — Wren AI carries it on paid plans only.
In your HTML
An analyst question Wren AI's MDL would resolve via a `signups` model and a `channel` dimension on your existing warehouse; nlqdb mints the Postgres database and answers the same English goal against the live schema in one element.
>
current month's signups grouped by acquisition channel <nlq-data goal="current month's signups grouped by acquisition channel"></nlq-data>Feature parity, honest
| Feature | nlqdb | Wren AI | Note |
|---|---|---|---|
| Owns the database (provisions + migrates) | |||
| Natural-language → SQL | |||
| Auto-migration via NL ('add a column for tags') | Wren AI sits on warehouses owned elsewhere; schema migration is not in its lane. | ||
| Semantic model / MDL with metrics + cubes | Wren AI's MDL (models, columns, relationships, views, cubes, metrics) is the differentiator over translator-only tools; nlqdb's contract is 'the live schema is the semantic layer'. | ||
| Row- and column-level access controls | Wren AI's RLAC/CLAC are part of the MDL primitive; nlqdb's per-DB `sk_live_*` keys scope by database, not by row or column. | ||
| Multi-engine support beyond Postgres | Wren AI's Apache DataFusion engine documents 22+ data sources including PostgreSQL, BigQuery, Snowflake, and DuckDB. nlqdb is Postgres-first in Phase 1; ClickHouse lands on the workload-analyser path. | ||
| MCP server with provisioning verbs | Wren AI ships a Python SDK, LangChain/LangGraph bindings, and skill bundles for Claude Code; no public MCP primitive for `create_database`. | ||
| Destructive-op diff preview before apply | Wren AI is a context layer over existing warehouses and does not perform DDL; diff preview before write/DDL apply is unique to nlqdb. | ||
| HTML embed element (in-product render) | |||
| SOC 2 Type II certification | Wren AI's public pricing page lists SOC 2 Type II on the Essential and Enterprise plans only; the Free plan lists no compliance bullets. nlqdb is pre-alpha and carries neither SOC 2 nor HIPAA yet. | ||
| Unlimited free-tier natural-language queries | Wren AI's Free plan documents a 20-monthly-credit allowance (plus 80 credits for the first 14 days). nlqdb's free chain (Groq → Gemini) is rate-limited per principal, not per question/month. | ||
| Open source / self-hostable | Wren AI is multi-licensed: Apache 2.0 covers `core/`, `sdk/`, `skills/`, `examples/`, and root files; `docs/` is CC-BY-4.0; future modules may land under AGPL-3.0-only per the LICENSE file. nlqdb is source-available on Functional Source License 1.1-ALv2 today; the LICENSE auto-converts to Apache 2.0 after two years (FSL 'ALv2 Future License' clause). |
shipped · partial · not shipped
Questions buyers ask
- How does nlqdb's NL→SQL compare to Wren AI's MDL-driven approach?
- Wren AI requires you to author a Modeling Definition Language file — models, metrics, cubes, relationships — that the LLM consults before writing SQL; the result is governed and reproducible but takes upfront semantic-modelling work. nlqdb prompts directly from the live Postgres schema fingerprint plus a recent-tables hint, with no MDL step. nlqdb publishes BIRD Mini-Dev + Spider 2.0-lite scores weekly to `docs/features/quality-eval/`; as of May 2026, Wren AI does not publish a single canonical accuracy benchmark.
- Can I point Wren AI at an nlqdb database?
- Yes — nlqdb provisions Postgres, and Wren AI lists PostgreSQL as one of its 22+ supported sources, so the connection string nlqdb returns slots in as a Wren AI data source. The reverse (nlqdb querying a Wren-AI-managed warehouse) is not supported today: nlqdb owns the database it queries, and bring-your-own-Postgres is on the roadmap rather than shipped.
- Does nlqdb ship a semantic layer like Wren AI's MDL?
- No. nlqdb's contract is 'the live schema is the semantic layer' — the schema fingerprint plus recent-tables hint is what the LLM sees, and `<nlq-data>` answers are written against that directly. If your team needs cubes, metrics, or row- and column-level access controls on top of the warehouse, Wren AI's MDL is the right shape and nlqdb does not replicate it today.
- Is nlqdb SOC 2 certified like Wren AI?
- No. nlqdb is pre-alpha; SOC 2 Type II is not in place yet. As of May 2026, Wren AI's public pricing page lists SOC 2 Type II on the Essential and Enterprise plans only — the Free plan does not carry it. If you're shipping a regulated product on a Wren AI paid tier today, that compliance posture is honest; nlqdb's compliance roadmap is downstream of GA.
- Can my AI agent provision a Wren-AI-managed warehouse the way it provisions an nlqdb database?
- Wren AI is a context layer plus agent toolkit — its Python SDK and LangChain bindings let an agent issue NL queries against an MDL-modelled data source, but the warehouse itself must already exist. nlqdb's MCP server (`mcp.nlqdb.com`) exposes `create_database`, `ask`, and `run`, so a Claude / Cursor / Cline agent stands up its own Postgres plus schema end-to-end without a human in the loop.
- Wren AI is open source — why not just self-host that?
- Pick Wren AI's OSS distribution if you want to self-host, audit, and modify the context layer against your own warehouse. The Canner/WrenAI repo is multi-licensed: Apache 2.0 covers `core/`, `sdk/`, `skills/`, `examples/`, and root files; `docs/` is CC-BY-4.0; future modules may land under AGPL-3.0-only per the LICENSE file. Pick nlqdb if you want the database, schema migrations, MCP-driven agent provisioning, and the HTML embed shipped as one product. nlqdb is source-available on Functional Source License 1.1-ALv2 today; the LICENSE auto-converts to Apache 2.0 after two years per the FSL 'ALv2 Future License' clause.
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 →