Canner/WrenAI

What WrenAI does

WrenAI is an open-source generative BI (GenBI) engine aimed at teams whose AI agents (Claude Code, Cursor, Cline, Codex, MCP clients, LangChain) need to answer business questions against a warehouse. Its core is the Modeling Definition Language (MDL): a Git-friendly definition of models, columns, relationships, cubes, metrics, enums, units and approved joins, stored as YAML and Markdown. Agents plan SQL against this semantic layer, validate it with a dry plan, and then execute it through a DataFusion-based engine spanning BigQuery, Snowflake, PostgreSQL, ClickHouse, Redshift, Databricks, DuckDB and others Connect a database.

The declared problem is trust: a bare agent sees schemas but not what the data means, so it writes plausible but wrong SQL. WrenAI adds a context layer (instructions.md, queries.yml, and a local LanceDB memory index) so definitions travel with the codebase. Beyond answers, agents can build browser-side dashboards via wren-core-wasm and deploy them to Vercel or Cloudflare Pages Build & deploy a GenBI app.

Requires a compatible agent environment; works with 50+ agents per the docs.

Quickstart

Prerequisite: Python with pip, and an agent that can follow skill instructions. Optionally a reachable database, or use the bundled jaffle_shop sample.

1. Install the CLI and any connector extras:

   pip install wrenai
   pip install "wrenai[postgres,memory]"
   

2. Install the agent skill in the project folder:

   npx skills add Canner/WrenAI
   

   This adds a short discovery stub; the agent fetches version-matched workflow guides from the CLI.

3. Ask the agent to set up the connection, for example: "Use Wren to set up my Postgres database." It creates a connection profile, scaffolds the project and runs a first query.

From there the documented loop is: enrich context from raw/ (wren skills get enrich-context), ask a business question so the agent writes and runs governed SQL, then ask for a filterable dashboard deployed to your Vercel or Cloudflare account (wren skills get genbi). See the full quickstart for details.

Usage constraints

The open-source engine (MDL semantic layer, governed text-to-SQL, MCP server, CLI, 22+ connectors) is Apache-2.0 and self-hostable open-core boundary. Row/column-level security, access control, the hosted GenBI UI, embedded APIs, scenario harnesses and audit/SLA features are commercial offerings in Wren Cloud or Enterprise Plus — plan for that if you need governed multi-user access.

Agent-driven dashboard deployment writes to your own Vercel or Cloudflare Pages account, so those credentials and quotas apply. The CLI supports connector extras installed selectively; memory uses a local LanceDB index. Users in mainland China may need a PyPI mirror and HF_ENDPOINT override for model downloads. The older chat-first product now lives on the legacy/v1 branch without new features or security fixes, so new adopters should use the current CLI/skill workflow.

HotRepo's take

WrenAI shows up on the daily GitHub all-language list in July 2026 with gains of 96 and 121 stars on consecutive days, and later at 52 stars in a single Python-list day on 2026-11-15, where it reached 15th on that language list. It does not have a recorded best rank on the all-language daily list, so the visible signal is mostly sustained Python-list activity rather than a top overall breakout. That pattern fits a project with a real engineering surface (Rust core, Python SDK, WASM dashboard runtime, agent SDKs) that keeps attracting developers as AI-agent BI becomes a common use case. The clear open-core split matters more than the star curve: the engine is genuinely open, while team governance and the hosted UI are paid, which is worth weighing before committing a production workflow.