RAG Converter at Heartland: Private Internal AI Without Uploading Documents

Signed by Heartland Industries•
A laptop on a farm-office desk with local document files open, no cloud upload in progress.

What is RAG Converter, and why is Heartland using it?

RAG Converter turns documents into AI-searchable chunks and embeddings. Heartland uses it on internal AI tools so people can ask questions of our own files — farm records, customer programs, compounding notes, LCA work — without uploading those files to a public model.

RAG means retrieval-augmented generation: the assistant looks up a passage from our corpus and answers from that, instead of guessing. The privacy move is where the conversion happens. RAG Converter's free converter is WebAssembly in the browser tab. Documents are read by the page. They are not sent to ragconverter.com to be parsed.

That is the same ownership rule we set in declouding customer and hemp supply-chain data: own the data you create, and do not trade it for a chatbot.

Why we will not paste Heartland files into a public LLM

Heartland's working set is not a blog post. It includes grower contracts, customer conversations, formulation notes, and trade-route detail we do not publish. Our farmers and manufacturer partners did not sign up to become training data.

Public chat tools are convenient. They are also a new outbound path. Once a PDF is in someone else's context window, you do not control retention, subprocessors, or whether a sentence shows up in another tenant's answer. NIST's Privacy Framework calls out processing on the local device as a way to limit observability and linkability (CT.DP-P1). Local conversion is that control, applied to RAG.

This is security as well as privacy. A leaked customer list and a leaked embedding of that list are both incidents. The difference is whether the first copy ever left the building.

How local conversion makes data more accessible

Declouding is easy to get wrong: lock the files in a drive nobody can search, and you have security theater plus slower work. Heartland's test is the opposite. If an engineer in Detroit and a grower liaison cannot find the same spec, the system failed.

RAG Converter's output is portable JSONL — one object per chunk — that loads into a vector store we operate. Same structure for a person dropping a folder in the tab and for a script hitting the convert API. Formats we actually have: PDF, Word, PowerPoint, Excel, Markdown, images, and audio. Scanned pages go through on-device OCR. There is no LibreOffice cloud queue and no "email us the file."

Performance. Drop-in for the people who already live in folders. No new desktop install for the browser path. Internal tools search the chunks we already produced.

Cost. The local converter is free and accountless because the work runs on the machine in front of you. We are not buying a second copy of our corpus from a SaaS RAG vendor.

Access. The point of embeddings is that a question can hit the right paragraph in a 200-page LCA pack or a grower handbook without forwarding the whole file. The published Imperium Farming LCA stays public. The drafts, emails, and customer overlays stay ours.

What stays local, and what we still put on a wire

Honesty about the split:

  • Browser converter. Files never leave the computer. The page may ask the server how many free conversions remain; that request is a number, not the document.
  • Developer API. POST /v1/convert is for pipelines and jobs the tab cannot run — full-resolution OCR, speech, video, no file-size cap. That path is a server. Heartland treats it as a tool we choose, not as a place to park the corpus. Convert, take the JSONL, store it where we already store Heartland data.
  • Cloud we have not killed yet. Identity, some customer systems, and email still live on vendors. The declouding plan is phased. RAG Converter is the piece we could pull local first: the conversion step that used to mean "upload the PDF to an AI product."

We do not claim the API is air-gapped. We claim the default Heartland habit is local conversion, and that internal AI answers from our index.

How this fits the rest of Heartland

This site already moved off WordPress to a static Rust stack for speed, SEO, and a smaller attack surface — the same Remade with Rust discipline we use on Heartland engineering work. The same instinct applies inside the company: keep the valuable files, make them searchable, do not rent that search to a model that wants the originals.

Partners who want the public story — Why Imperium, Imperium Filler, Imperium Fibers — still get it here. The files behind those pages stay on our side of the wire.

Try the converter at ragconverter.com. For how Heartland thinks about customer and supply-chain data, read the declouding article or contact us. More of the library: sustainability news.

Quick answers

About this article

What is RAG Converter at Heartland?
Heartland uses RAG Converter so internal AI can search company documents without uploading them. Local conversion, portable chunks, and data we still own.
What is RAG Converter, and why is Heartland using it?
RAG Converter turns documents into AI-searchable chunks and embeddings. Heartland uses it on internal AI tools so people can ask questions of our own files - farm records, customer programs, compounding notes, LCA work - without uploading those files to a public model.
What makes we will not paste Heartland files into a public LLM?
It includes grower contracts, customer conversations, formulation notes, and trade-route detail we do not publish. Our farmers and manufacturer partners did not sign up to become training data.
How local conversion makes data more accessible?
Declouding is easy to get wrong: lock the files in a drive nobody can search, and you have security theater plus slower work. If an engineer in Detroit and a grower liaison cannot find the same spec, the system failed.
Where do I talk to Heartland about RAG Converter at Heartland?
The company path is /why-imperium and the contact form is /contact.

Contact

Get In Touch

Talk to our materials science team — replies within one business day.

Heartland office locations: Detroit, New York, and Paris.DetroitNew YorkParis
Get in touch