RagBytes API

Base URL

https://api.ragbytes.com

Try it

Each tier has its own endpoint. Upload the file to it as multipart/form-data in a field called file. Pick a tier and a document below, copy the code, or send the request straight from this page.

curl -F "[email protected]" https://api.ragbytes.com/v1/chunk/basic

Chunking endpoints

Basic and Structured take the same upload and return the same response. Supported file types: .pdf, .docx, .html, .htm, .md, .markdown, .txt.

• POST/v1/chunk/basicBasic. About 2,000 characters per chunk with a 200-character overlap. Small paragraphs under one heading are packed together, and long tables split between rows.
• POST/v1/chunk/structuredStructured. Chunks follow the document's heading tree. Each section becomes one chunk where it fits, small neighbouring sections are merged under their shared heading, and long sections are split evenly.
• POST/v1/chunk/semanticComing soonSemantic. Compares the meaning of neighbouring passages with embeddings and starts a new chunk where the topic shifts.
• POST/v1/chunk/enrichedComing soonEnriched. Adds a summary, keywords and document context to every chunk, and describes images and diagrams.

Other endpoints

• GET/v1/methodsList the available tiers, the endpoint for each, and the file types the service accepts.Run
• GET/v1/statsSee how many documents have been chunked and how many chunks were made. Only these two totals are kept.Run
• GET/healthCheck the service is up and see which build it is running.Run

Response

Chunks come back in document order. For each one, send embed_text to your embedding model and store content and metadata next to the vector.

• embed_text is the heading path, a blank line, then the chunk's text. The heading path tells the embedding model where the chunk sits in the document.
• content is the chunk's text, or base64 bytes when metadata.media_type is an image type.
• metadata.format is one of text, list, table, image, chart, code or complex.
• metadata.page is the page number for PDFs, otherwise null.

{
  "filename": "guide.md",
  "file_id": "6f1c…",
  "chunks": [
    {
      "chunk_id": "b2e4…",
      "metadata": {
        "filename": "guide.md",
        "file_id": "6f1c…",
        "chunk_number": 0,
        "headings": ["Guide", "Install"],
        "heading": "Guide",
        "subheading": "Install",
        "page": null,
        "format": "text",
        "media_type": null,
        "details": { "start": 0, "end": 20 }
      },
      "embed_text": "Guide > Install\n\nInstall the package.",
      "content": "Install the package."
    }
  ]
}

Errors and limits

Errors return JSON with a detail message.

• 400The file could not be read.
• 404No such tier, or the tier isn't available yet.
• 413The upload is over 50 MB.
• 415The filename suffix isn't a supported file type.
• 429More than 10 chunking requests in 10 seconds from your address. Wait 10 seconds and retry.
• 524The document took longer than about 100 seconds to process. Try splitting it.