Real AI agent use cases, with sources and setup prompts
How to scan and organize receipts and paperwork with AI (ChatGPT, Paperless-ngx)
Short answer: For a quick sort, scan receipts with your phone (iPhone Notes or Files: Scan Documents; Android: Google Drive's Scan button) into PDFs, upload them to ChatGPT or Claude and ask for a table of date, vendor, total, tax, category and payment method that you save as a CSV. To keep every document searchable for years, run Paperless-ngx: free, open-source software that runs in Docker on your own computer, reads every scan with OCR, files it by type, tag and sender, imports attachments from your email, and since version 3 can suggest titles and tags and answer questions with a local Ollama model, so nothing has to leave your house. Keep the originals the IRS asks for (generally 3 years after filing, longer in some cases).
Real setups: a Grok Bot that files snapped receipts into a yearly Google Sheets book, a bookkeeping bot that matched a month of receipts overnight, and an agent that files insurance claims when the receipt email lands. 11 matching use cases below, each with a public source.
Prices are from each official page, checked Oct 27, 2026; they exclude tax and can change. Swipe the table sideways on a phone.
Turn a pile of receipts into a spreadsheet with ChatGPT or Claude
About 15 minutes for a month of receipts. Free plans work for small batches.
Scan with your phone. iPhone: open Notes, tap the attachment (paperclip) button > Scan Documents, let it auto-capture each page, tap Keep Scan and Done, then share the PDF. Android: open Google Drive, tap the camera (Scan) button, capture, crop, tap Done and save as PDF. Lay receipts flat on a dark surface in good light.
Name the columns you want. Decide the columns once and reuse them every month. This header works in Excel, Numbers and Google Sheets.
Upload and extract. Upload up to about 20 receipts at a time to a new chat (or a Claude Project with your categories saved) and paste the first prompt below. Ask for a downloadable CSV.
Check the numbers. Ask for a second table of anything it was unsure about (blurry totals, missing tax, foreign currency), compare those with the paper, then check that the column total matches your card statement for the same dates.
File the originals. Rename each PDF to the file_name it suggested (date_vendor_total.pdf) and keep them in one folder per year, for example Receipts/2026, in iCloud Drive, Google Drive or OneDrive.
Run Paperless-ngx at home for every document
About 30 minutes. Needs Docker; works on a Mac, Windows PC, Linux box or NAS. Documents stay on your computer.
Option A: the official install script. Mac users need GNU sed and wget first (brew install gnu-sed wget). The script asks where to store files, your time zone, OCR language and an admin login.
shell
# Docker must already be installed (Docker Desktop on Mac/Windows)
bash -c "$(curl --location --silent --show-error https://raw.githubusercontent.com/paperless-ngx/paperless-ngx/main/install-paperless-ngx.sh)"
# answers a few questions, then starts everything; open http://127.0.0.1:8000
Option B: download the compose files yourself. The SQLite version is the simplest: one Paperless container plus a small broker.
shell
mkdir paperless && cd paperless
B=https://raw.githubusercontent.com/paperless-ngx/paperless-ngx/main/docker/compose
curl -L -o docker-compose.yml "$B/docker-compose.sqlite.yml"
curl -L -o docker-compose.env "$B/docker-compose.env"
curl -L -o .env "$B/.env"
# make a secret key and paste it into docker-compose.env
python3 -c "import secrets; print(secrets.token_urlsafe(64))"
Edit docker-compose.env. PAPERLESS_SECRET_KEY must be changed from change-me. The default date order is day-month-year, so US users should set MDY.
text
PAPERLESS_SECRET_KEY=paste-the-long-random-string-here
PAPERLESS_TIME_ZONE=America/New_York
PAPERLESS_OCR_LANGUAGE=eng
PAPERLESS_DATE_ORDER=MDY
PAPERLESS_CONSUMER_RECURSIVE=true
PAPERLESS_CONSUMER_SUBDIRS_AS_TAGS=true
# Linux only: match the owner of ./consume (check with: id -u ; id -g)
# USERMAP_UID=1000
# USERMAP_GID=1000
Start it and import.
shell
docker compose pull
docker compose up -d
# first visit asks you to create the admin account
open http://localhost:8000 # Linux: xdg-open, Windows: start
# drop files in ./consume to import them; consume/Receipts/2026 tags them Receipts + 2026
You can also drag files onto the web page or upload from your phone's browser. Each file is OCR'd, so its text becomes searchable.
Set up types, tags and email. In the web app create document types (Receipt, Invoice, Bill, Contract, Tax, Warranty), tags such as tax-2026, unpaid and car, and correspondents for frequent senders. For bills that arrive by email, open Settings > Mail, add your mail account (Gmail and Outlook can sign in with OAuth), then add a rule that imports PDF attachments from a folder or sender and tags them; mail is checked every 10 minutes.
Find anything later. Search uses words in the document plus filters. Add a Monetary custom field called Total to receipts if you want to sort by amount.
text
type:receipt created:[2026-01-01 to 2026-12-31]
tag:tax-2026 correspondent:"home depot"
type:invoice tag:unpaid
added:[-1 week to now]
warranty tag:electronics
Add private AI tagging and Document Chat with Ollama
Uses the AI features built into Paperless-ngx v3. Runs on your own machine; plan on 8 GB or more of free RAM.
Add an Ollama container. Docker Compose reads docker-compose.override.yml automatically, so you don't have to edit the official file.
yaml
# save as docker-compose.override.yml next to docker-compose.yml
services:
ollama:
image: ollama/ollama
restart: unless-stopped
volumes:
- ollama:/root/.ollama
webserver:
depends_on:
- ollama
volumes:
ollama:
Turn on the AI settings. llama3.1 is Paperless-ngx's default model for Ollama; embeddinggemma powers search for Document Chat. To use a cloud model instead, set PAPERLESS_AI_LLM_BACKEND=openai-like with PAPERLESS_AI_LLM_API_KEY, and remember document text is then sent to that provider.
text
# add to docker-compose.env
PAPERLESS_AI_ENABLED=true
PAPERLESS_AI_LLM_BACKEND=ollama
PAPERLESS_AI_LLM_ENDPOINT=http://ollama:11434
PAPERLESS_AI_LLM_MODEL=llama3.1
PAPERLESS_AI_LLM_EMBEDDING_BACKEND=ollama
PAPERLESS_AI_LLM_EMBEDDING_MODEL=embeddinggemma
The document index for chat is rebuilt by a daily background task; new documents become chat-searchable after it runs.
Use the suggestions. Open a document: the suggestion buttons now include AI ideas for title, date, tags, correspondent and type. To apply them automatically, create a Workflow with the trigger Document Added and the action Apply AI Suggestions.
Ask your documents. Use Document Chat from the toolbar for questions such as 'When does the dishwasher warranty end?' or 'What did we pay the plumber in 2026?' Check the answer against the document it links to.
Copy-paste prompts
Written for this setup. Replace anything in [brackets] before you send it.
1. Extract a batch of receipts into a CSV
After uploading receipt photos or PDFs.
These are my receipts for [month]. For each one, extract: date (YYYY-MM-DD), vendor, short description, subtotal, tax, tip, total, currency, payment method (last 4 digits only), category from this list [Groceries, Dining, Fuel, Home repair, Medical, Kids, Office, Travel, Other], and business or personal.
Return a CSV I can download with the header: date,vendor,description,subtotal,tax,tip,total,currency,payment_method,category,business_or_personal,file_name. Make file_name date_vendor_total.pdf.
Don't guess: if a number is unreadable, leave it blank and list that receipt in a separate 'check by hand' table.
2. Build a tax-time summary
Once a year, from your receipts CSV.
Here is my receipts CSV for [year]: [paste or upload].
Group the business expenses by category with totals, list medical expenses separately, and flag anything that looks personal in the business list or duplicated (same vendor, amount and date). Give me a short list of what to ask my tax preparer. Don't tell me what's deductible as fact; mark it as a question.
3. Find a missing receipt in my email
An agent with email access (Grok Bot, Claude with Gmail, Codex).
Search my email from [date range] for receipts, invoices and order confirmations from [vendor] around [amount].
For each match, give me the date, sender, amount and subject, save the PDF or a printout of the email to [folder] named date_vendor_total.pdf, and list any card charges in this CSV [upload statement] that still have no receipt. Don't reply to or delete any email.
4. Design my Paperless-ngx setup
Before importing hundreds of documents.
I'm setting up Paperless-ngx for [household of 2 adults, a small side business, a rental property]. The documents are: [receipts, utility bills, bank and card statements, tax forms, insurance, medical, car, warranties, kids' school letters].
Suggest at most 8 document types, 15 tags and the correspondents I should create, a consume subfolder layout (subfolders become tags), and a PAPERLESS_FILENAME_FORMAT that stores files by year and type. Keep it simple enough that I'll actually use it.
5. Write a Paperless-ngx mail rule
Bills and receipts that arrive by email.
I get these by email: [Amazon order receipts, the electric company's monthly bill PDF, my insurance statements]. Write Paperless-ngx mail rules for each: which folder to watch, the filter (sender, subject or attachment name), whether to import attachments only, which document type, tags and correspondent to assign, and what to do with the email afterwards (mark read or move). Explain each field in plain words.
6. Check a bill for mistakes
Medical, phone, utility or repair bills.
Here is a bill: [upload]. List every line item with its amount, add them up and check the total, and flag duplicate charges, fees that look new compared with last month's bill [upload], and anything I should ask about. Then write a short, polite message I can send to dispute the items you flagged.
7. Build a warranty and returns tracker
After scanning receipts for electronics and appliances.
From these receipts [upload], find anything that usually has a warranty or a return window (electronics, appliances, tools, furniture). Make a table: item, store, purchase date, price, return deadline if the receipt shows one, manufacturer warranty length if printed, and the date it ends. Mark any length you're assuming rather than reading from the receipt, and sort by the soonest deadline.
Safety checklist
Black out full account numbers, Social Security numbers and dates of birth before uploading statements or tax forms to a chatbot.
In Paperless-ngx, a cloud AI backend sends document text to that provider. Use the Ollama backend if your documents must stay at home.
Change PAPERLESS_SECRET_KEY from change-me, use a strong admin password with 2FA, and don't expose port 8000 to the internet; use Tailscale for remote access.
AI can misread totals and dates, especially on faded thermal receipts. Check extracted numbers against the original before filing taxes or expense reports.
Back up: Paperless-ngx's exporter writes all documents and metadata to ./export; copy that folder to another drive or cloud storage regularly.
Keep paper or original electronic copies as long as the IRS or your tax authority requires; a spreadsheet summary alone is not a receipt.
Family Book is a Grok Bot template: snap a receipt or let it read order emails, and it files receipts in your cloud storage and keeps the yearly income and expense book in Google Sheets or Excel Online, in any currency and language.
Her bookkeeping bot Peg worked through April overnight, matching receipts and searching personal accounts and cards for possible business expenses, following rules they'd written together and listing questions for her to review in the morning.
When a receipt from his massage therapist arrives by email, his agent logs into the insurance site, files the claim, stores the receipt, watches for the "claim processed" email and adds each claim to a dated spreadsheet with receipt links for tax time, for him, his wife and his kids.
He gives Claude Code and Codex access to the Xero API, Xero in the browser for reconciliation, his email to find documents, Stripe and GoCardless APIs, past tax returns and a CPA portal, plus the Google Workspace CLI to create documents.
He built an app in Codex that pulls receipts from his email into SQLite. Hermes uses that purchase history to predict what needs restocking, then logs into his supermarket every week, books a delivery slot and fills the cart. He can photo a barcode to add an item.
She set up a Grok Bot called Peg to do her bookkeeping on a voice call, using the QuickBooks Online connector from Grok Bot's connector list. Peg enters invoices and bills, records payments and matches them to invoices.
He rebuilt Claude Duet's main features in Grok Bot in about 10 minutes, using this prompt: "Every day at 7:00 send me a morning brief: today's calls, emails that need my answer, overdue invoices, and the one thing to do first. Draft replies, but send nothing without my approval." He also set up inbox sorting, calendar moves around a focus block and invoice reminders.
A builder open-sourced a business app for freelancers that keeps clients, quotes, invoices and messages on your own machine with no cloud. The AI proposes and you decide, and it ships with a Claude Code skill and a free book.
Sultan, which sells grooming products to barbershops and salons, runs a Grok Bot called Finance-Erik for bookkeeping, bank reconciliation, invoices, VAT and cash position. It matches transactions, prepares bookings and chases missing receipts, but never pays anything; price changes, tax calls and big reclassifications go to the owner.
How do you scan and organize receipts and paperwork with AI (ChatGPT, Paperless-ngx)?
For a quick sort, scan receipts with your phone (iPhone Notes or Files: Scan Documents; Android: Google Drive's Scan button) into PDFs, upload them to ChatGPT or Claude and ask for a table of date, vendor, total, tax, category and payment method that you save as a CSV. To keep every document searchable for years, run Paperless-ngx: free, open-source software that runs in Docker on your own computer, reads every scan with OCR, files it by type, tag and sender, imports attachments from your email, and since version 3 can suggest titles and tags and answer questions with a local Ollama model, so nothing has to leave your house. Keep the originals the IRS asks for (generally 3 years after filing, longer in some cases).
What's the easiest way to organize receipts with AI?
Scan them with your phone's built-in scanner into PDFs, upload a batch to ChatGPT or Claude and ask for a table of date, vendor, total, tax and category, then download it as a CSV for your spreadsheet. Check the totals against the paper before you rely on them.
What is Paperless-ngx?
A free, open-source document manager you run yourself, usually in Docker. It OCRs every scan so the text is searchable, files documents by type, tag and correspondent, pulls attachments from email, and in version 3 can suggest metadata and chat about documents with a local or cloud AI model.
Is it safe to upload receipts and tax documents to ChatGPT?
Receipts are usually low risk, but bank statements, tax forms and medical bills carry account numbers and ID numbers. Black out those numbers first, turn off training on your data in the app's settings, or keep the documents local with Paperless-ngx and Ollama.
Can Paperless-ngx use AI without sending documents to the cloud?
Yes. Set PAPERLESS_AI_LLM_BACKEND=ollama and point it at an Ollama server on your own network. Its docs warn that with a cloud backend, document content is sent to that provider.
How long should I keep receipts for taxes?
The IRS says generally 3 years from when you filed, 6 years if you left out more than 25% of your income, 7 years for a bad-debt or worthless-securities deduction, and records for property until the time limit runs out for the year you sell it. Ask a tax professional about your case.
Can AI read faded or crumpled receipts?
Often, if the scan is sharp and well lit, but thermal receipts fade and models can misread totals. Flatten them, scan in good light and compare the extracted total with the paper; Paperless-ngx keeps the original image next to the OCR text.
Which AI tools do people use to scan and organize receipts and paperwork with AI (ChatGPT, Paperless-ngx)?
Across the 11 sourced use cases in this guide, the most used tools are Grok Bot, Claude Code, OpenAI Codex, bank accounts, Claude Duet.
How much does it cost to get started?
1 of the 11 use cases can start on a free plan. The most common paid starting price is $30/mo (6 use cases). Prices come from each tool's official pricing page.
How long does setup take?
3 of the 11 use cases list a setup time of an hour or less. Each card shows the setup time the builder reported.
Are these real examples?
Yes. Every use case links to its public source (an X post, a Hacker News thread or a GitHub repo) with the author and date, so you can check it yourself.
The Daily 5
Liked this guide? Get five new use cases tomorrow morning
Five new real AI use cases every morning. Free. Unsubscribe any time.