OCR component

The ocr_image component extracts text from an image using tesseract - fully local, no API key, no network call. It is a kdeps component from the components asset set: call it with a component: resource, no install step. The component is a thin client: the tesseract call itself is declared in the ocr_image tool asset and runs in the kdeps API.

Where it runs

Both workflow mode and agent mode. In agent mode the same tool is the ocr_image built-in, a declared exec tool; the component and the agent both reach it through the kdeps API.

Requirements

The tesseract binary on PATH (apt install tesseract-ocr, brew install tesseract). When it is missing, kdeps installs tesseract-ocr from the tool asset's exec.install on the next run, and kdeps build bakes it into the image.

Basic usage

component: ocr_image -> POST http://kdeps.api/apis/tools.kdeps.io/v1/tools/ocr_image/call
                    -> tools/ocr_image.yaml exec: tesseract <file> stdout -l <language> --psm <psm> --oem <oem>
                    -> .stdout = the text

# resources/ocr.yaml
actionId: ocr
name: Extract Text
component:
  name: ocr_image                  # from the components asset set; a components/ocr_image/ of your own wins
  with:
    file: /data/scanned-invoice.png  # required

# resources/respond.yaml
actionId: respond
requires: [ocr]
apiResponse:
  success: true
  response:
    text: "{{ output('ocr') }}"   # the component output is the extracted text

Inputs

Input	Default	Description
file	(required)	Path to the image: png, jpg/jpeg, tiff, bmp, gif. Text-layer PDFs go through the load_document component instead
language	eng	tesseract -l value, e.g. deu, eng+fra. Needs the matching tesseract-ocr-<lang> data pack
psm	3	Page segmentation mode (--psm, 0-13); 3 is fully automatic, 6 assumes one uniform block of text
oem	3	Engine mode (--oem, 0-3); 3 uses whatever engine is available

component:
  name: ocr_image
  with:
    file: /data/receipt.jpg
    psm: 6   # a single uniform block of text

Output

The extracted text, as a string:

"kdeps OCR Sample\n"

A missing tesseract binary fails the call: the API returns 400, and the component's failOnError: true stops the workflow.

The component itself

kdeps update components/ocr_image refreshes it. To change how it runs, copy it into your project's components/ocr_image/component.yaml - a local component with the same name wins:

# components/ocr_image/component.yaml (as shipped, abridged)
apiVersion: kdeps.io/v1
kind: Component
metadata:
  name: ocr_image
interface:
  inputs:
    - {name: file, type: string, required: true}
    - {name: language, type: string, default: eng}
    - {name: psm, type: integer, default: 3}
    - {name: oem, type: integer, default: 3}
  output: "{{ output('call').body }}"   # the tool returns the text
resources:
  - actionId: call
    httpClient:
      method: POST
      url: http://kdeps.api/apis/tools.kdeps.io/v1/tools/ocr_image/call
      headers: {Content-Type: application/json}
      failOnError: true
      timeout: 60s
      data:
        file: "{{ get('ocr_image.file') }}"
        language: "{{ get('ocr_image.language') }}"
        psm: "{{ get('ocr_image.psm') }}"
        oem: "{{ get('ocr_image.oem') }}"

The tesseract command line lives in tools/ocr_image.yaml, the same tool the agent loop uses:

# assets/tools/ocr_image.yaml (as shipped, abridged)
parameters:
  file: {type: string, required: true}
  language: {type: string, default: eng}
  psm: {type: integer, default: 3}
  oem: {type: integer, default: 3}
exec:
  command: tesseract
  install: tesseract-ocr
  paths: [file]
  args:
    - "{{.args.file}}"
    - stdout
    - "-l"
    - "{{.args.language}}"
    - "--psm"
    - "{{.args.psm}}"
    - "--oem"
    - "{{.args.oem}}"

See also

• Component resource - how component: calls pass with: inputs
• Transcribe components - the equivalent pattern for audio
• Resources overview - all resource types