file-processor

Demonstrates the file input source - a single-shot workflow that reads a document from a file and returns a structured AI analysis.

Structure

file-processor/
├── workflow.yaml
└── resources/
    ├── 01-read-file.yaml    # expose file metadata
    ├── 02-summarize.yaml    # LLM summarization
    └── 03-response.yaml     # structured JSON output

Usage

# Pass a file path via --file flag (highest priority)
kdeps run examples/file-processor --file /path/to/document.txt

# Pipe content via stdin
cat document.txt | kdeps run examples/file-processor

# Use an environment variable
KDEPS_FILE_PATH=/tmp/doc.txt kdeps run examples/file-processor

# Pipe JSON with inline content
echo '{"path":"/tmp/doc.txt","content":"Your text here"}' | kdeps run examples/file-processor

Requirements

• Ollama running locally with llama3.2 installed (ollama pull llama3.2)

Input Resolution Priority

1. --file <path> CLI argument
2. stdin (raw text or JSON {"path":"...","content":"..."})
3. KDEPS_FILE_PATH environment variable
4. input.file.path config field

Accessing File Data

Inside resources, use the input() expression:

Expression	Value
input('fileContent')	The file's text content
input('filePath')	The source file path
input('content')	Alias for fileContent
input('path')	Alias for filePath

The file Source

settings:
  input:
    sources: [file]
    file:
      path: ""   # optional default path; override via --file or stdin

The workflow runs once and exits after processing the file.

Sample Output

{
  "file": "/path/to/document.txt",
  "analysis": {
    "title": "Introduction to Machine Learning",
    "summary": "This document covers the fundamentals of ML...",
    "key_points": [
      "Supervised learning uses labeled training data",
      "Neural networks are inspired by the human brain",
      "Regularization prevents overfitting"
    ]
  }
}

Validate

kdeps validate examples/file-processor