Carmenta — OCR In Pure Rust Without Tesseract Or Python

OCR is how a computer reads the world that was never a PDF. A scanned invoice. A settings screen. A whiteboard. The default stack was Tesseract — C++, GPL gravity, a deploy footprint that looks like an operating system — or a cloud API that takes the page and calls it "processing." Python wrappers made both feel modern. Neither belongs next to a vault.

Carmenta is OCR in Rust: documents, screens, and a change-gated LIVE frame reader. Detection, recognition, reading order from geometry. No Python, no C/C++ by default, nothing GPL. The FFAI crate is ffai-carmenta. The catalog is Remade with Rust. The thesis is rebuilding the foundation of the internet. This article is the reader.

Why Carmenta Is OCR In Rust Without Tesseract

Tesseract is a fine research lineage and a terrible mesh dependency. GPL at the bottom of the stack gates who can ship. A C++ engine on a wasm32 tab is a science project. Cloud OCR means the page left the building. The NIST Secure Software Development Framework pushes memory-safe languages for this class of parser. Carmenta is that push. OCR in Rust means the hot path is Rust.

ffai-carmenta is how FFAI exposes Carmenta. Same engine, media-library surface. Pull Carmenta when reading is the product. Pull FFAI when reading sits next to speech and detection.

Trust, Security, and Incentive fail when "we read it locally" still means Tesseract or a hosted API. Trust fails when the invoice boarded a vendor. Security fails when the engine is C++. Incentive fails when the safer path is GPL or huge. Carmenta is how those three primitives apply to a page.

What Carmenta Ships As ffai-carmenta

github.com/Remade-With-Rust/carmenta is the source. Documents and screens are the strength. Wild street photography is the acknowledged trade-off — the right trade for edge OCR. ffai-carmenta is the crate in the FFAI graph. Carmenta is OCR in Rust you can actually deploy.

Detection, Recognition, Reading Order In Carmenta

OCR is three jobs. Find the text. Read the text. Put the lines in the order a human would. Carmenta does all three, and reading order comes from geometry — not from hoping the model guessed left-to-right. That is how a two-column invoice stays an invoice.

OCR in Rust that only dumps a bag of words is a demo. Carmenta is the pipeline.

Change-Gated LIVE Frames In Carmenta

The LIVE reader is change-gated. A camera pointed at a screen should not burn the CPU on identical frames. Carmenta runs when the pixels moved. That is the doorbell / kiosk / shop-floor path. ffai-carmenta is how that path sits next to Diana and Mercury in FFAI.

AI on the edge is the pipeline article. remade_ffmpeg_rs opens the frame. Diana boxes. Carmenta reads. Mercury hears.

No Python, No C/C++ By Default, Nothing GPL

Those three refusals are the product. Carmenta is not a Tesseract binding. ffai-carmenta is not a Python wheel with a Rust CLI. Nothing GPL means a startup can embed OCR in Rust without a license review that lasts a quarter. The EFF has argued for decades that a free internet needs infrastructure you can embed. Carmenta is that argument for text in pixels.

The Electronic Frontier Foundation has documented how "the document stayed on device" claims collapse when the helper still uploads the page. Carmenta keeps the page.

How Carmenta Feeds The Distributed Cloud

A distributed cloud that cannot read a document on the node uploads the contract. Carmenta is OCR in Rust so the node reads it.

Trust The Page, Secure The Reader

Trust is a page that stayed yours. Security is Carmenta — ffai-carmenta, memory-safe, no C++ by default. Incentive is OCR in Rust with no GPL and a footprint measured in megabytes of detector weights, not gigabytes. The NIST Zero Trust Architecture tells operators to stop trusting the perimeter. Carmenta is how you stop trusting the OCR API as well.

The Freedom Guide is local-first steps. Invoices in that guide that still hit a cloud reader are a contradiction. Carmenta ends it.

Disco Party, RAG, And Era-Shaped Documents

Disco Party places work products on the mesh. RAG Converter already runs Carmenta client-side. Era 1 stores the file. Era 3 is work and money — the pages that run a life. Era 4 is a home companion that reads a label on a screen.

The FTC keeps catching products that collected more than they needed "to make search work." OCR in Rust on the device deletes that excuse.

The Catalog Entry For Carmenta

Source is github.com/Remade-With-Rust/carmenta. The FFAI surface is ffai-carmenta. Cards and shipping status live on the Remade with Rust catalog. The mesh this reader feeds is what a distributed cloud is and the Disco Party. Digital Freedom is Trust, Security, and Incentive.

If you came from the Learn index looking for OCR in Rust, this is Carmenta. If you came from Tesseract, this is the remake — no Python, no GPL. The internet you want to own still has pages. Carmenta is how that sentence stays true without the C++.

The detector beside this reader is Diana. Office XML without libxml2 is rusty_xml. Keeping that work off a vendor model is local AI.