Fetch Files from URLs in Jinba Flow

Choose a file input tool to fetch files from a URL and pass them to later workflow steps.

Choose a file input tool by the kind of URL or file your flow receives, then pass its output to the next step.

Choose a file input tool

The File Input tools allow you to fetch files from specified URLs.

Key Features

• INPUT_PDF_FILE: Fetch PDF files from URLs.
• INPUT_CSV_FILE: Fetch CSV files from URLs.
• INPUT_DOCX_FILE: Fetch DOCX files from URLs and return the extracted text.
• INPUT_JSON: Input JSON data directly.
• INPUT_FILE: Input generic files from URLs.
• INPUT_TEXT: Input plain text data directly.

Tools

INPUT_PDF_FILE

Fetch PDF files from URLs for processing.

Input:

• value (string, required): URL of the PDF file to fetch

Output:

• File content and metadata for further processing

INPUT_CSV_FILE

Fetch CSV files from URLs and parse them into rows.

Input:

Parameter	Type	Required	Description
value	string / string[]	Yes	URL(s) of the CSV file(s) to load. Can be a single URL or an array of URLs
description	string	No	Description of the CSV file
encoding	string	No	Character encoding of the file: auto (default, detected automatically), utf8, utf-8, latin1, iso-8859-1, shift_jis, euc-jp, gbk, big5, or windows-1252

Output:

• result: the parsed rows as an array of string arrays (one inner array per row, including the header row). When value is an array of two or more URLs, result is an array of such row arrays, one per file (a one-element array collapses to a single row array).

CSV text is parsed with RFC 4180 semantics: a quoted field may contain commas and line breaks, and "" inside a quoted field decodes to a literal ". A leading BOM is stripped, blank lines are skipped, and rows with a different number of fields are kept as-is rather than rejected. A quoted field that is never closed fails the step with a ValidationError that names the record where the problem starts, instead of silently dropping the rest of the file.

Use this tool when later steps need CSV rows or cells. If characters come out garbled, set encoding (for example shift_jis). See Summarize a CSV file for a transformation example.

INPUT_DOCX_FILE

Fetch DOCX (Word) files from URLs and return the extracted text.

Input:

Parameter	Type	Required	Description
value	string / string[]	Yes	URL(s) of the DOCX file(s) to load. Can be a single URL or an array of URLs
description	string	No	Description of the DOCX file

Output:

• result: Text extracted from the DOCX file(s)

INPUT_JSON

Input JSON data directly as a string.

Input:

• value (string, required): JSON string to parse

Output:

• Parsed JSON data ready for processing

INPUT_FILE

Input generic files from URLs.

Input:

Parameter	Type	Required	Description
value	string / string[]	Yes	URL(s) of the file(s) to fetch. Can be a single URL or an array of URLs
description	string	No	Description of the file
file_extension	string	No	Expected file extension (e.g. pdf). Separate several with commas to accept any of them (e.g. xlsx,xlsm). Known extensions are validated against the fetched file’s MIME type; extensions without a registered MIME type (e.g. jsonl) are validated against the fetched filename.

Output:

• result: the fetched file URL as a string (an array of URLs when value was an array). Pass it to a downstream step that takes a file URL; the file body is not read into the output.
• filenames / extensions: the filename(s) and extension(s) extracted from the URL(s), in the same shape as result.

INPUT_TEXT

Input plain text data directly.

Input:

• value (string, required): Plain text content

Output:

• Text content for processing

Examples

Example: Fetch Files from URLs

- id: "pdf"
  tool: INPUT_PDF_FILE
  input:
    - name: value
      value: https://www.w3.org/WAI/ER/tests/xhtml/testfiles/resources/pdf/dummy.pdf

- id: "csv"
  tool: INPUT_CSV_FILE
  input:
    - name: value
      value: https://people.sc.fsu.edu/~jburkardt/data/csv/addresses.csv

- id: "docx"
  tool: INPUT_DOCX_FILE
  input:
    - name: value
      value: https://example.com/files/report.docx
    - name: description
      value: "Monthly report Word document"

- id: "json_data"
  tool: INPUT_JSON
  input:
    - name: value
      value: '{"source": "example", "items": [1, 2, 3]}'

- id: "config_file"
  tool: INPUT_FILE
  input:
    - name: value
      value: https://raw.githubusercontent.com/example/repo/main/config.yaml
    - name: description
      value: "Application configuration file"

- id: "text_input"
  tool: INPUT_TEXT
  input:
    - name: value
      value: "This is sample text content for processing"

Example: Summarize a CSV file

Load a sales CSV, total the amount column per region in Python, then ask a model for a report.

- id: sales_csv
  tool: INPUT_CSV_FILE
  input:
    - name: value
      value: "https://example.com/sales_data.csv"
    - name: description
      value: "Monthly sales CSV with region and amount columns"

- id: summarize_sales
  tool: PYTHON_SANDBOX_RUN
  needs: [sales_csv]
  input:
    - name: data_type
      value: OBJECT
    - name: code
      value: |
        import json

        rows = json.loads(r'''{{steps.sales_csv.result}}''')
        header, records = rows[0], rows[1:]
        amount_col = header.index("amount")
        region_col = header.index("region")

        totals = {}
        for row in records:
            if len(row) <= max(amount_col, region_col) or not row[amount_col].strip():
                continue
            region = row[region_col]
            totals[region] = totals.get(region, 0) + float(row[amount_col].replace(",", ""))

        print(json.dumps({
            "row_count": len(records),
            "total_sales": round(sum(totals.values()), 2),
            "sales_by_region": {k: round(v, 2) for k, v in totals.items()},
        }))

- id: sales_report
  tool: OPENAI_INVOKE
  needs: [summarize_sales]
  input:
    - name: prompt
      value: |
        Write a short markdown sales report with an executive summary,
        key metrics, and recommendations, based on these totals:
        {{steps.summarize_sales.result.data}}

{{steps.sales_csv.result}} is inserted as JSON text, so json.loads returns the rows as a list with the header first; cells are strings, so convert numbers yourself. The printed JSON becomes steps.summarize_sales.result.data (for example result.data.total_sales), and the report text is in steps.sales_report.result.content. Rename amount and region to match your file’s header row.

Example: JSON Data Processing

- id: user_data
  tool: INPUT_JSON
  input:
    - name: value
      value: |
        {
          "name": "John Doe",
          "age": 30,
          "email": "[e-mail]",
          "preferences": {
            "theme": "dark",
            "notifications": true
          }
        }

- id: process_user_data
  tool: OPENAI_INVOKE
  needs: [user_data]
  config:
    - name: version
      value: gpt-5.2
  input:
    - name: prompt
      value: |
        Process this user data and create a personalized greeting:
        {{steps.user_data.result}}
        
        Create a friendly greeting message that includes their name and references their preferences.

Download limits

The file input tools (INPUT_FILE, INPUT_PDF_FILE, INPUT_CSV_FILE, INPUT_DOCX_FILE, INPUT_IMAGE_FILE, INPUT_FILE_AS_TEXT) apply these limits to every URL they fetch:

Limit	Value
URLs per step (when value is an array)	50
Files downloaded at the same time	4
Time per download, including the body	5 minutes
Size per file	100 MiB; 20 MiB for INPUT_IMAGE_FILE and INPUT_DOCX_FILE

A step that exceeds a limit fails without downloading the rest of the file, and when one file of a list fails, the downloads still in progress are stopped. INPUT_CSV_FILE and INPUT_IMAGE_FILE also fail when the URL answers with an HTTP error status, instead of reading the error page as data.

Related Input Tools

The following input tools are documented on other pages:

• INPUT_WEBHOOK, INPUT_SLACK_EVENT, INPUT_IMAGE_FILE, INPUT_FILE_AS_TEXT, INPUT_TWILIO_SMS: see Choose an input tool
• INPUT_JSON_WITH_VALIDATION: see Input JSON with Validation