I built an agent that works while I sleep. It found a pipe leak, a price hike, and a subscription I forgot I had.

Description and use

Napper is an offline autonomous agent that ingests CSV exports, forecasts with TabPFN‑3.5, screens for anomalies, audits transactions, and delivers a concise morning brief, catching price hikes, duplicate charges, electricity spikes, and a water pipe burst.

Problem

The author wanted to stop manually checking multiple CSV exports each morning and only be notified when something important happened.

Outcome

On demo data Napper detected four anomalies: a gym price hike, a duplicate power bill charge, a 3‑day electricity spike, and a water pipe burst, plus a forgotten subscription. No false alarms were reported.

Workflow

• Drop CSVs into a local folder
• Napper ingests and normalizes data, hashing to avoid duplicates
• TabPFN‑3.5 generates 30‑day forecasts with holdout validation
• Model‑free seasonal screen compares today to median of previous eight same‑weekdays
• Recurring‑charge detector flags price hikes and duplicate charges
• Morning brief is generated and sent via text and TTS

Evidence

I generated four synthetic CSVs (bank, electricity, steps, and water) with real‑world storylines planted in them, seeded so anyone can reproduce the exact results.

Source

Four catches across four distinct detectors: Gym Membership Hike, Double Billing, 3-Day Electricity Spike, Water Pipe Burst, Forgotten Subscription.

Source

TabPFN-3.5, an open-weight tabular foundation model from Prior Labs that runs on your local CPU/GPU, to generate recursive 30-day forecasts with holdout validation.

Source