Breaking

Open-source app ReceiptHero uses Llama 4 Scout 17B for receipt OCR

July 5, 2026 at 08:26 EDT

Open Source
Foundation Models
Software Dev & Coding

ReceiptHero, an open-source receipt management app that uses Meta's multimodal Llama 4 Scout 17B as its OCR engine to automatically extract vendor, date and amount from receipt images, has been released on GitHub. Built by developer Nutlope, the MIT-licensed app is available in its GitHub repository, with demos and app screens shown on the dedicated site usereceipthero.com.

Open Source · MIT License

ReceiptHero: Reading Receipts with an LLM, Not OCR

A privacy-first app that drops the traditional OCR pipeline and hands receipt images straight to Meta's Llama 4 Scout — extracting vendor, date, amount and items, then auto-classifying spending, all stored locally in your browser.

17B

active parameters at inference

109B

total parameters (MoE)

experts, natively multimodal

MIXTURE OF EXPERTS · ACTIVE vs TOTAL PARAMETERS

17B

Active

109B

Total

Only ~16% of the model's parameters fire per inference — efficiency of MoE at Scout's scale.

HOW IT WORKS

Drop receipt image

→

Llama 4 Scout reads it via Together AI

→

Extract fields + classify

→

Stored locally in browser

STRENGTHS

Native LLM image understanding — no Tesseract
Clear, spelled-out workflow in the README
Local storage = privacy-focused by design

HURDLES

Requires a Together AI API key to run locally
Some setup — a modest barrier for beginners

The Stack

Llama 4 Scout 17B Next.js (App Router) Tailwind CSS shadcn/ui + Radix UI Zod Together AI MIT License

358 stars

40 forks

68 commits

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.