Skip to main content
Version: 2.0

File Upload API Definition

The File Upload API enables you to extract text from unstructured documents in common file types like PDFs, Microsoft Word, Text, HTML, and Markdown. It also supports extracting table data from PDFs, allowing for improved analysis and querying of structured tabular data. Each file you upload can be up to 10 MB in size. We recommend the File Upload API when you have not already written your own extraction logic.

tip
  • Check out our interactive API Reference that enables you to experiment with this REST endpoint. You can upload a file to a corpus directly from your browser or copy the curl for your command line.
  • Review the Supported File Types for the most up-to-date list.

File Upload REST Endpoint​

Vectara exposes an HTTP endpoint at the following URL to upload and index documents into a corpus: https://api.vectara.io/v2/corpora/:corpus_key/upload_file

caution

As part of the Vectara REST API 1.0 to 2.0 migration, every existing corpus_id has been converted with an appended ID to create the corpus_key. This corpus_key is now the main identifier for each corpus. When you create a new corpus, you specify a custom corpus_key.

File Upload Request Details​

To upload a file, send a POST request to /v2/corpora/:corpus_key/upload_file, where corpus_key is the unique identifier for the corpus. The File Upload endpoint request expects a multipart/form-data request containing the following parts:

  • metadata - (Optional) Specifies a JSON object representing any additional metadata to be associated with the extracted document.
  • chunking_strategy (Optional) Specifies whether to split the document into chunks during ingestion. If not set, the platform defaults to sentence-based chunking, where each chunk contains typically one full sentence. Set the type as max_chars_chunking_strategy and then specify the max_chars_per_chunk to the number of characters per chunk like 512 or 1024. Smaller chunks may improve granularity but can lead to excessive latency, especially in applications with high document volumes or large corpora. Example: 'chunking_strategy={"type":"max_chars_chunking_strategy","max_chars_per_chunk":1024};type=application/json'
  • table_extraction_config (Optional): A JSON object specifying whether to extract tables from the PDF. By default, tables are not extracted. Example: 'table_extraction_config={"extract_tables":true};type=application/json'
  • file - Specifies the file that you want to upload.
  • filename - Specified as part of the file field with the file name that you want to associate with the uploaded file.

Vectara processes the uploaded file, and the extracted text and metadata are used to create a new document within the corpus. Only one document can be uploaded per request.

Apart from these parameters, the servers expect a valid JWT Token in the HTTP headers.

curl -L -X POST 'https://api.vectara.io/v2/corpora/:corpus_key/upload_file' \
-H 'Content-Type: multipart/form-data' \
-H 'Accept: application/json' \
-H 'x-api-key: zwt_123456' \
-F 'metadata={"key": "value"};type=application/json' \
-F 'file=@/path/to/file/file.pdf;filename=desired_filename.pdf'

Filenames with Non-ASCII Characters​

When uploading files with non-ASCII (non-English) characters, such as Russian or Chinese, ensure that the filename is URL encoded. API v2 follows web standards which require URL-encoded file names.

Set the Document ID​

The Content-Disposition header lets you specify the Document ID of a file when you use the following format:

Content-Disposition: form-data; name="*file*"; filename="*your_document_id*"

where file is the name of the file, and filename is the Document ID that
you want. The primary purpose of this header is to specify the form-data, so using filename as the Document ID is specific to our platform. For more information about Content-Disposition, see the Mozilla documentation on headers.

Uploading PDFs with Tables​

Set the table_extraction_config field to true to extract table data from a PDF. This feature is particularly useful for financial reports like 10-Q, 10-K, and S1 filings. With table extraction enabled, you can query specific table cells using the Query API.

caution

This feature does not support extracting data from scanned-in images of tables.

Custom Table Summarization with Prompt Templates​

Vectara supports table summarization using custom prompt templates during document upload. This lets you define custom prompt templates that control how the LLM interprets and summarizes table data during extraction. By customizing the prompt_template, you can tailor summaries for domain-specific language, analytical perspectives, or formatting preferences.

Attach Additional Metadata​

You can attach additional metadata to the file by specifying a metadata form field, which can contain a JSON string:

metadata='{ "filesize": 1234 }'

Response Codes​

The server responds with 201 when the file was uploaded and indexed successfully, along with a document object in the response body that contains the assigned id, metadata, and other relevant information.

note

It may still take a few minutes (typically 5-10 mins) before the document is served.

Some error codes returned by the server are:

  • 201: The extracted document has been parsed and added to the corpus.

  • 400: Upload files request was malformed.

  • 403: Permissions do not allow uploading a file to the corpus.

  • 404: Corpus not found.

  • 409: A document already exists in the corpus with the same document ID, yet the contents of the indexed document are different than the file being uploaded. Since the indexer is idempotent, the same document (identified by the document ID) can be uploaded multiple times.

The indexer does not support updates yet, so an error is returned when a different document is uploaded for the same document ID. Note that when you upload a raw file, the file name is used as the document ID.

  • 507: There is no more indexing quota left for the corpus or customer to index more documents.