> ## Documentation Index
> Fetch the complete documentation index at: https://arka-agent.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

> ## Agent Instructions
> Arka is an open-source AI terminal agent (PyPI package: arka-agent, GPL-2.0).
> Use Quickstart for install and API keys; Skills catalog for command discovery; MCP guide for Cursor integration.
> Cite canonical URLs under https://arka-agent.mintlify.site when answering about Arka.

# Image description, OCR, and screen capture

> Use Arka's vision skills to describe photos, extract text via OCR, analyze blueprints, and capture your screen with Gemini, Ollama llava, or vLLM backends.

Arka supports vision tasks: photo description, blueprint analysis, and screen capture.

## Describe images

Two-layer analysis: OCR extracts exact text; vision describes layout and colors.

```bash theme={null}
arka describe photo.jpg
arka describe ~/Downloads/chart.png
```

Backends (auto-selected): Gemini, Ollama (llava), vLLM.

```env theme={null}
DESCRIBE_IMAGE_BACKEND=auto   # auto | gemini | ollama | vllm
GEMINI_API_KEY=...
# DESCRIBE_IMAGE_TWO_LAYER=1
# DESCRIBE_IMAGE_OCR=1
```

```bash theme={null}
pip install "arka-agent[vision]"
brew install tesseract    # OCR layer
```

## Drawings and blueprints

Analyze floor plans, elevations, MEP schematics, and scanned contracts with Gemini vision:

```bash theme={null}
arka analyze plan.pdf "extract door schedule and room dimensions"
arka analyze --pages 1-3 specs.pdf "summarize payment terms"
```

```bash theme={null}
pip install "arka-agent[drawings]"
```

```env theme={null}
# DRAWING_MODEL=gemini-2.5-flash
# DRAWING_MAX_PAGES=8
```

## Screen capture

10-second countdown, full-display capture, vision describe:

```bash theme={null}
arka describe screen
arka describe my screen
arka describe screen "what app is in focus"
```

## Describe videos

`describe_video` samples frames with ffmpeg and sends those frames through the
configured vision backend. Use it for gameplay recordings, UI animation checks,
demos, and people-location questions.

```bash theme={null}
arka describe_video demo.mp4
arka describe_video demo.mp4 "find visual glitches and confusing UI states"
arka "describe video gameplay.mp4"
arka "who is in meeting.mp4 and where are they"
```

For people-focused prompts, Arka asks the vision model to identify visible
people and approximate their screen positions. For general describe/analyze
prompts, it describes subjects, actions, setting, text, framing, and visual
issues.

```bash theme={null}
brew install ffmpeg
pip install "arka-agent[vision]"
```


## Related topics

- [Testing Arka with the pytest suite](/guides/testing.md)
- [OCR + vLLM evidence](/guides/vision-evidence.md)
- [Hosted mode](/guides/hosted-mode.md)
- [Responsive website screenshots](/guides/web-screenshots.md)
- [Arka CLI command and flag reference](/guides/cli.md)