Custom Prompts

💡TL;DR

Notemd lets you override the default LLM prompt for every task type. Custom prompts give you control over what the model prioritizes: you can constrain concept extraction to a specific discipline, change the linking style, or redirect translation tone. Prompt variables like {{content}} and {{concept}} are substituted at runtime. The focused learning domain feature prepends your field of study to automatically scope extraction results.

This is part of the Obsidian AI Knowledge Management Guide.

Overview

Notemd's built-in prompts are written for general-purpose knowledge work. They work well across domains, but they also leave room for domain-specific improvement. Custom prompts let you tell the LLM exactly how to behave for your use case -- whether that means extracting only medical terminology, linking only named entities, or translating in a formal register.

Custom prompts do not replace the entire system prompt. They override or augment the task-specific instruction portion while preserving the structural output format (e.g., CONCEPT: lines, wiki-link syntax) that Notemd depends on for parsing.

How It Works

Prompt Architecture

Every Notemd task follows the same prompt structure:

[System role: You are a helpful assistant...]
[Domain constraint: Relevant Fields: [{domain}]]   <-- focused learning (optional)
[Custom task prompt]                                <-- your override (optional)
[Output format instructions]                        <-- always preserved
[Content: {{content}}]                              <-- runtime substitution

1. Base system prompt -- Sets the assistant persona. Not overridable.
2. Focused learning domain -- Prepends Relevant Fields: [medicine / law / computer science / ...] to constrain scope. This is a settings toggle, not a raw prompt.
3. Custom task prompt -- Replaces the default task instructions (e.g., what kinds of concepts to extract, how aggressively to link).
4. Output format -- Structural rules (CONCEPT: lines, wiki-link format). Always appended by Notemd to ensure parseability.
5. Content -- The actual note text, substituted at runtime via {{content}}.

Prompt Variables

Variables are replaced at execution time. Available variables differ by task:

Variable	Tasks	Replaced With
{{content}}	All	The note or selection text
{{concept}}	Concept note template	The extracted concept name
{{description}}	Concept note template	LLM-generated concept description
{{backlinks}}	Concept note template	List of source notes with links
{{date}}	Concept note template	Current date (YYYY-MM-DD)
{{language}}	Translation	Target output language
{{source}}	Concept note template	Source note basename

Focused Learning Domain

The focused learning feature is a shortcut for domain scoping. Instead of writing a custom prompt that mentions your field, you set focusedLearningDomain in settings:

Settings → Advanced → Focused Learning Domain
Example: "medicine" or "computer science" or "Chinese philosophy"

This prepends Relevant Fields: [medicine] to the extraction prompt, causing the LLM to prioritize domain-relevant concepts and suppress generic ones. It is lighter than a full custom prompt and works across all extraction tasks.

Configuration

Setting	Default	Effect
customAddLinksPrompt	''	Override the default add-links task prompt
customExtractConceptsPrompt	''	Override the default concept extraction prompt
customResearchPrompt	''	Override the default research + summarize prompt
customTranslatePrompt	''	Override the default translation prompt
customDiagramPrompt	''	Override the default diagram generation prompt
focusedLearningDomain	''	Prepend discipline context to extraction and linking prompts
extractConceptsMinimalTemplate	true	Use title-only concept note template (affects {{concept}} template scope)
customConceptNoteTemplate	''	Override the concept note file template with full variable support

When a custom prompt field is empty, the built-in default prompt is used.

Example

You are a medical researcher extracting concepts from clinical trial notes. The default prompt extracts generic nouns alongside medical terms, which clutters your concept folder.

Solution: Set a focused learning domain and a custom extraction prompt.

1. Set focusedLearningDomain to medicine
2. Set customExtractConceptsPrompt to:

Extract only clinically relevant concepts: diseases, drug names,
anatomical terms, measurement endpoints, and statistical methods.
Ignore generic English nouns, brand names, and institutional names.
Prioritize MeSH terminology where applicable.

Now when you run "Extract concepts" on a clinical note, the LLM produces:

CONCEPT: Randomized Controlled Trial
CONCEPT: Double-Blind Method
CONCEPT: Myocardial Infarction
CONCEPT: Creatine Kinase-MB
CONCEPT: Kaplan-Meier Estimate

Instead of the default behavior that might also extract unwanted generic terms like "hospital" or "month".

Tips

• Start with the focused learning domain before writing a full custom prompt. It covers 80% of use cases with a single setting.
• Preserve output format -- Never remove format instructions like "one per line" or "CONCEPT: prefix" from custom prompts. Notemd's parser depends on them.
• Test with a single note first -- Custom prompts can have unexpected effects on extraction breadth. Verify on one note before batch processing.
• Keep prompts concise -- Overly long custom prompts consume tokens and can dilute the model's focus.

---

Next Steps

• Wiki-Links -- What gets linked when you customize the add-links prompt
• Concept Notes -- Concept note templates and the {{concept}} variable
• Batch Processing -- Apply custom prompts across entire folders
• Troubleshooting -- Fix issues where custom prompts produce unexpected output