Transactions

A transaction is a single financial event — an expense, a purchase, or an income entry. Ozzie uses GPT-4o to extract structured transaction data from natural language, images, PDFs, and spreadsheets.

What a transaction contains

Field	Type	Description
id	UUID	Ozzie-assigned identifier
amount_cents	integer	Amount in cents (always positive)
currency	string	ISO 4217 code (e.g. USD, BRL, EUR)
category	enum	Spending category (see below)
description	string	Short label extracted by AI (max 60 chars)
transaction_date	date	YYYY-MM-DD — extracted from input or defaults to today
source	enum	How the transaction was submitted
ai_confidence	float	AI confidence score from 0.0 to 1.0
created_at	datetime	When Ozzie received it

Categories

Ozzie classifies every transaction into one of 10 categories:

Category	Examples
food	Groceries, restaurants, coffee, delivery
transport	Gas, Uber, bus, parking, car maintenance
housing	Rent, mortgage, home repairs
utilities	Electricity, internet, water, phone bill
health	Doctor, pharmacy, gym, dental
entertainment	Streaming, movies, games, concerts
education	Courses, books, tuition, school supplies
clothing	Clothes, shoes, accessories
income	Salary, freelance, dividends, refunds
other	Anything that doesn't fit the above

Input sources

Transactions can come from 6 different sources:

Source	Channel	Input format
api_text	Direct API	Natural language string
api_image	Direct API	Base64-encoded image (receipt/statement)
api_pdf	Direct API	Extracted text from a PDF
api_spreadsheet	Direct API	CSV rows as plain text
whatsapp_text	WhatsApp webhook	Message text from Evolution API
whatsapp_image	WhatsApp webhook	Receipt photo from Evolution API

AI parsing

When you submit "Spent $45 on groceries at Whole Foods and $5 on coffee", Ozzie returns two transactions — not one. GPT-4o extracts every financial event mentioned.

The ai_confidence score tells you how sure the AI was:

Score	Meaning
0.9 – 1.0	High confidence — clear amount, category, and date
0.7 – 0.89	Good — some inference was needed (e.g. currency from context)
0.5 – 0.69	Moderate — ambiguous input, may want to confirm with user
< 0.5	Low — could not extract reliably; consider re-prompting

How transactions fuel insights

Every stored transaction feeds into:

• Chat context — the AI coach knows the user spent $X on food this month
• Plan tracking — actual spend vs budgeted allocations
• Goal progress — did they save/pay-down what the plan said?
• Spending trends — category breakdowns over time (accessible via GET /transactions)

→ See Transactions API Reference