Configuration

All settings are available under Settings > Transcriber.

Ollama Configuration

Setting	Type	Default	Description
Server URL	text	http://localhost:11434	The URL of your Ollama server
Test connection	button	—	Verifies the Ollama server is reachable and refreshes models
Vision model	select	qwen3.5:9b	Dropdown populated from installed Ollama models
Recommended models	list	—	Install buttons for recommended models not yet installed
Install custom model	text	—	Enter any Ollama model name to install it

Model management

Models can be managed in two ways:

1. Command palette — Use the Install AI model, Select AI model, and Remove AI model commands for quick access without opening settings. See Usage for details.
2. Settings tab — The sections below provide the same functionality within the settings UI.

The Vision model dropdown is populated dynamically from models installed in Ollama. When you open settings, the plugin queries Ollama for installed models. If the currently selected model is no longer installed, it appears in the dropdown with a “(not found)” warning.

The Recommended models section shows vision models known to work well for transcription that are not yet installed:

• maternion/LightOnOCR-2:1b
• qwen3.5:2b
• qwen3.5:4b
• qwen3.5:9b — default
• qwen3.5:27b
• qwen3.5:35b

Click Install next to any recommended model to download it directly. A progress notice shows download status.

You can also install any Ollama model by entering its name in the Install custom model field and clicking Install. This is useful for quantized variants (q4_K_M, q8_0, bf16) or other vision-capable models.

Once installed, the model is automatically selected and appears in the dropdown.

Transcription Settings

Setting	Type	Default	Description
Transcription prompt	text	(see below)	The prompt sent to the vision model along with each image
Include subfolders	toggle	off	When transcribing a folder, also process images in subfolders
Overwrite existing files	toggle	off	Overwrite existing .md files when re-transcribing images

Default transcription prompt

The default prompt tells the model it is an expert document transcriber and asks it to convert the image to Obsidian-flavored Markdown. Specifically, it instructs the model to:

• Preserve all text content exactly as shown
• Use appropriate Markdown formatting (headings, lists, tables, code blocks, links, etc.)
• Maintain the original document structure and hierarchy
• Transcribe handwritten text as accurately as possible
• Convert diagrams or charts to Mermaid syntax, describing them in detail if necessary
• Use [unclear] as a placeholder for illegible text
• Output only the transcribed Markdown — no comments, explanations, preamble, code fences, or framing text

You can customize this prompt to suit your specific use case (e.g. focus on handwriting, technical diagrams, or specific languages).