When should you use it?
- You avoid finance apps that want your messages
- Your bank sends an email per transaction
- You have a small always-on computer
Not for: Anyone without bank alert emails; Shared mailboxes
What do you need, and what do you get?
You need
- Mailbox with bank and merchant emails
- Raspberry Pi or similar
- Local language model
You get
- Nightly categorized spending log
- Combined totals for every bank and card
How does it work, step by step?
- Turn on alertsMake sure every account emails you per transaction.
- Have the agent build the parserAsk a coding agent to write the email reader and categorizer.
- Run a local modelKeep categorization on your own device.
- Match merchant emailsAttach order or receipt details to each charge.
- Schedule nightlyRun by cron and review the summary weekly.
The prompt
Our own prompt, inspired by Fahad Hasin and written to work in Claude Code and Custom agent. Values in lime are placeholders.
Write a Python script that runs nightly on my Raspberry Pi. It should read new emails from my bank alert senders, extract amount, date, account and merchant, look for a matching merchant email from the same day to add item details, categorize each charge with the local model at [endpoint], and append rows to spending.csv. Never send data to an external service. Print a daily summary by category.
Inspired by Fahad Hasin: #40: How I am using Claude Code as a non-coder. The prompt above is written by playbooks.bot in our own words; nothing from the original is quoted.
What can go wrong?
- Bank email formats change; check the parser when totals look odd.
- A small local model mislabels some merchants; correct and add rules.
Our take
Privacy is the real feature here. Running on a home device with a local model means spending data need not leave the house. Fahad's setup answers a common objection to finance apps: he did not want a company reading his text messages, so the alerts are read and categorized on a Raspberry Pi at home by a local model. The trade-off is upkeep. Banks change their email formats, so if a total looks odd, check the parser first. A small local model mislabels some merchants; correct those and add rules rather than switching to a cloud model by default. Matching merchant receipts to charges makes the ledger far more useful than raw alert lines. Give the reader access only to the bank alert label, encrypt the Pi's storage, and review the weekly summary instead of trusting it blindly. This only works if every account you hold sends an email for each transaction.
Where the time goes back: Expense tracking happens without manual entry or giving a company access to your inbox.
FAQ
Why build instead of using an app?
Did it work?
Source and credit
Inspired by Fahad Hasin. We describe the idea in our own words, link to the original, and wrote the prompt ourselves; nothing is quoted.
- Inspired by
- #40: How I am using Claude Code as a non-coder
- Creator
- Fahad Hasin
- Published on
- fahadhasin.substack.com (Substack)
- Original date
- Mar 3, 2026
- License
- Not openly licensed (the poster keeps all rights), so we quote nothing and link to the original
- Originally for
- Claude Code
- AI dry run
- Oct 23, 2026: passed. Dry-run tested by our AI test agent on sample input. PASS: Gates held on desk check; no live sends or purchases.
AI dry run on sample data; not run with live connected accounts. The playbooks.bot test agent filled the prompt with synthetic inputs, produced the output and checked it against the page's stated outputs and guardrails. - Verified
- Oct 24, 2026: source link resolves to the original, the credited creator published it, our excerpt or prompt matches it, and the license allows our use.
Are you the creator? Claim or correct this page.