Ask the data

Ridgeline's AI analyst is a Claude agent with six read-only tools over this warehouse: search the dbt catalog, describe a model, trace lineage, list the governed metrics, query a metric through MetricFlow, and run a guarded SELECT. It answers the way an analytics engineer would: find the right mart, prefer the governed metric, and cite where the number came from.

This page shows a recorded run of the 20-question eval rather than a live chat box, so it costs nothing to visit. Each answer was checked against a value computed straight from the marts.

Run on Oct 30, 2026 , with data as of Oct 29, 2026 . The rest of the site rebuilds every night, so today's numbers may differ slightly from these answers.

The 20 questions

Open a question to see the agent's full reply, the tools it called, and the governed metric or SQL behind the number. The last one asks for something the warehouse doesn't hold, and the right reply is "no data", not a guess.

Try it yourself

The same tools run as an MCP server, so you can point Claude Code or Claude Desktop at a local copy of the warehouse:

git clone https://github.com/sky-five/ridgeline-analytics && cd ridgeline-analytics
python scripts/run_all.py                                   # build the warehouse
claude mcp add ridgeline -- python -m analyst.server        # then ask Claude about Ridgeline

The guardrails live in analyst/sql_guard.py: the database opens read-only, file and network access are turned off, every statement must parse as a single SELECT, and rows and run time are capped.