ragconverter.com vs LlamaParse: local files and an API
ragconverter.com and LlamaParse both prepare documents for retrieval. LlamaParse parses a file you upload and can index it inside LlamaCloud. RAG Converter can do the conversion in the browser so the file never uploads, and the same conversion is an API that returns chunked embeddings as JSONL.
How does ragconverter.com compare with LlamaParse?
ragconverter.com can convert a document in the browser so the file never uploads, and the same job is an API that returns chunked embeddings as JSONL. LlamaParse is a hosted parser: you upload the file and receive markdown, text, or JSON, with indexing available inside their platform.
Two products sitting at different points in the pipeline
LlamaParse is the document platform from LlamaIndex. Parse takes an uploaded file and returns markdown, text, or JSON. The same account can Extract fields into a schema you define, and Index the result so retrieval stays inside LlamaCloud. ragconverter.com is a converter. A document becomes passages, each with its text, an embedding, and metadata such as the source file, page, and heading path, written as JSONL you load into whatever vector database you already run.
Localization is whether the bytes move
LlamaParse's ordinary path sends the document to their service. An enterprise VPC is still their boundary, arranged on their plan, and the file still leaves the laptop to reach it. The browser path on ragconverter.com parses, chunks, and embeds in the tab. A contract, a patient note, or an unreleased draft can be converted with no upload, which is the form of localization a privacy policy cannot replicate, because there is no copy on a server to retain or breach. The constraint, and why it matters for the files people actually convert, is covered in why convert documents locally.
The API exists for the jobs a tab should not do
The free tab is budgeted for a browser: 8 MB per file and 400 chunks, inside a 4 GB WebAssembly address space that a large scanned page can fill. A signed-in tab raises that budget and is still a tab. Nightly jobs, video, and OCR at full resolution belong on a server. The API is a multipart POST to /v1/convert. It returns the same chunk objects the download produces, and the manifest names the embedding model plus any file that was skipped. A folder dropped into the page and a script that posts the same folder share one importer. The shape of that contract is in a RAG conversion API for developers.
Where LlamaParse is the better parser
Agentic OCR for charts, handwriting, and ugly scans is the job LlamaParse sells, and it is a real job. A converter that reads a text layer and says so when a page has none is doing a different thing from a model that reads the pixels. If the documents that matter are the worst fifty in the pile, run those through both and keep the parse a person can still read. Retrieval follows that sample. A leaderboard published by either vendor does not.
How to choose for a product
Use the browser when the person holding the file should not have to upload it. Use the API when conversion has to run unattended and the material is allowed to move for the length of the request. Use LlamaParse when you already live in LlamaIndex and the hard part is their parser, then embed with the model your index expects. The vectors ragconverter.com ships are the ones its manifest names. The passage text sits in the same object when you need a different encoder. Both the tab and the API run on Remade with Rust libraries, so the two runtimes are one conversion with two budgets.