Agentic data analysis · GroundFree to use · Claude Code · Pi

Evidence before action.

Statistical search across your whole dataset, ranked into evidence an agent can act on and a human can audit.

$ npm install -g @ai-outfitter/outfitterthen pick the Data Analyst profile

Artifact

194 million rows. One overnight run.

From the published HHS Medicaid analysis: the finder's own event log. The report's highest-scoring pattern: a segment of organizationally complex, high-volume, low-cost services with mean spending growth of 179,156% — explosive growth from small bases, delivered with the segment conditions attached.

run · hhs-medicaid · finder events

06:53:47 data loaded 194,002,693 rows · 10.33 GB · 17.5s
06:54:10 search started KPI: TOTAL_PAID · 40 conditions · depth 5
07:30:24 patterns found 9,588 candidates · 2,174s · 30 workers
12:37:22 patterns ranked 300 kept by KS score · 0 errors, 0 warnings
Unsupervised.com — March 2026

How it works

Search, don’t guess

An agent exploring by SQL tests one hypothesis per query. It only finds what someone thought to ask. Finder searches the space itself: 100x more data at 1/100th the cost of query-by-query exploration.

Ranked, with receipts

Every pattern arrives ranked by statistical strength, with the segment conditions visible — never a black box. Caveats the system found are recorded alongside.

Built for agents

The output is evidence an agent can act on: ranked candidates, open questions, and a trail a human can audit before anything ships.

Quickstart

Run it on your own data today.

Try it in Pi

terminal

$ npm install -g @ai-outfitter/outfitter
$ outfitter # pick the Data Analyst profile

The Data Analyst profile bundles the Finder workflows and sets up the Finder binary the first time you ask for a pattern search.

Try it in Claude Code

terminal

$ curl -fsSL https://unsupervised.com/cli/install.sh | sh
$ unsupervised finder --help

For warehouse-connected, team-scale use, see Finder for Teams.