Wren AI and PlyDB share a lot on the surface: both are open source under Apache 2.0, both are read-only, and both connect AI agents to databases via native MCP servers. The difference is the layer in between. Wren AI translates natural language into SQL through a governed semantic model (MDL) — the goal is letting business users ask questions without writing SQL. PlyDB gives AI agents direct SQL access with no translation layer — the goal is letting agents that already write SQL reach live data as simply as possible.
| Wren AI | PlyDB | |
|---|---|---|
| Natural language to SQL | Built-in — Wren AI translates natural language to SQL through its MDL semantic layer | Any AI agent connected to PlyDB handles natural language and writes the SQL directly |
| Semantic context | MDL — engineer-authored model defines entities, metrics, and business logic upfront; enforced consistently across all queries | OSI overlays — agents discover schema automatically and write context incrementally across sessions; no upfront modeling required |
| Charts & report generation | Auto-generated charts, dashboards, and spreadsheet exports alongside query results | Not applicable — returns query results; visualization is handled by the agent or downstream tools |
| Business user self-service | Natural language UI built-in — business users ask questions directly without an agent | Any AI agent (Claude, GPT, Gemini) can serve as the natural language interface — no SQL knowledge required from the user |
| Agent integration (MCP) | Native MCP — Wren Engine exposes the semantic layer to Claude, Cline, Cursor, and other agents | Native MCP & CLI — direct SQL access to configured sources |
| Direct SQL access | Queries route through the MDL semantic layer — suited to governed, modeled data | Agents construct and execute arbitrary SQL directly against live sources |
| Cross-source queries | Single source per project — cross-source requires a Trino integration as a workaround | JOIN across any connected source in one query — no workaround needed |
| Read-only | Read-only by design | Read-only by design |
| Open source | Apache 2.0 | Apache 2.0 |
| Cost | Free tier (20 credits/month); paid plans from $99/month; enterprise on request | Open source — Apache 2.0 |
Wren AI is an open-source Generative BI agent built around a governed semantic layer. Data engineers define their data model in MDL (Model Definition Language) — a JSON format that encodes entities, metrics, dimensions, and business rules — and Wren AI uses that model to translate natural language questions into accurate SQL. Business users ask questions; Wren AI writes and executes the SQL and returns results alongside auto-generated charts and dashboards. A native MCP server exposes the semantic layer to AI agents via Claude, Cline, and Cursor. Wren AI connects to PostgreSQL, MySQL, BigQuery, Snowflake, DuckDB, Clickhouse, and more — but each project is scoped to a single data source; cross-source queries require a Trino integration. Wren AI Cloud is available for teams who don't want to self-host.
PlyDB is an open-source gateway built from the ground up for AI agents that already write SQL. You declare your data sources in a single JSON config file — PostgreSQL, MySQL, SQLite, S3, files, Google Sheets — and any AI agent connects immediately via native MCP or CLI, with no semantic translation layer in the way. PlyDB's semantic context system works differently from MDL: it auto-discovers schema and provides an OSI-format overlay system where agents themselves record institutional knowledge — enum meanings, business rules, domain context — that persists and compounds across sessions. No upfront modeling phase; agents build understanding incrementally as they query. Read-only by design, single binary, minutes to first query.
These tools can complement each other. Wren AI handles the human-facing layer — business users asking natural language questions against a governed semantic model. PlyDB handles the agent layer — giving AI agents direct SQL access to live operational databases, flat files, and sources that don't belong in a semantic model. Both are read-only, both are Apache 2.0, and both can run alongside each other in the same stack.
Your agent writes the SQL. PlyDB connects it to your data. No modeling phase. No credits. Open source.
Apache License 2.0