Distill Memory

Capture breakthrough moments and valuable insights as searchable memories in your knowledge base.

When to Suggest

Breakthrough moments:

• Extended debugging finally resolves
• User relief signals (“Finally!”, “Aha!”, “That was it!”)
• Root cause discovered after investigation

Important decisions:

• Compared multiple options
• Chose with clear rationale
• Trade-off resolved with reasoning

Research conclusions:

• Investigated multiple approaches
• Reached definitive conclusion
• Optimal path determined

Unexpected discoveries:

• Counterintuitive solution found
• Assumption challenged and corrected
• Surprising cause-effect relationship

Lessons learned:

• “Next time do X instead”
• Preventive measure identified
• Pattern recognized for future use

Skip these (not worth preserving):

• Routine fixes
• Work in progress
• Simple Q&A
• Generic information

Prerequisites

nmem CLI – Choose one option:

Option 1: uvx (Recommended)

curl -LsSf https://astral.sh/uv/install.sh | sh
uvx --from nmem-cli nmem --version

Option 2: pip

pip install nmem-cli
nmem --version

Ensure Nowledge Mem server is running at http://localhost:14242

Usage

Use nmem m add to create memories:

nmem m add "Insight content with context for future use" \
  -t "Searchable title (50-60 chars)" \
  -i 0.8

Memory Quality Guidelines

Good memories (atomic + actionable):

• “React hooks cleanup must return function. Missing return caused memory leaks in event listeners.”
• “PostgreSQL over MongoDB: ACID compliance needed for financial transactions.”
• “Docker build cache invalidation: COPY package*.json before source files.”

Poor memories (avoid):

• Vague: “Fixed bugs in the code”
• Too long: Full conversation transcripts
• No context: “Use useState”

Content Guidelines

• Focus on outcome/insight, not process
• Include “why” not just “what”
• Add enough context for future understanding
• Be specific and actionable

Importance Scores

Options

Available unit types: fact, preference, decision, plan, procedure, learning, context, event.

Suggestion Approach

Timing: After resolution/decision, when user pauses

Pattern: “This [type] seems valuable – [essence]. Distill into memory?”

Frequency: 1-3 per session typical. Quality over quantity.

Examples

# High-value debugging insight
nmem m add "React hooks cleanup must return function. Missing return caused memory leaks in event listeners when component unmounted." \
  -t "React Hooks Cleanup Pattern" \
  -i 0.9 --unit-type learning -l react -l frontend

# Architecture decision
nmem m add "Chose PostgreSQL over MongoDB: needed ACID compliance for financial transactions and complex JOIN queries for reporting." \
  -t "Database Choice: PostgreSQL for ACID" \
  -i 0.9 --unit-type decision -l database -l architecture

# Development workflow tip
nmem m add "Docker build cache: COPY package*.json and run npm install BEFORE copying source files. Saves rebuild time on code changes." \
  -t "Docker Build Cache Optimization" \
  -i 0.7 --unit-type procedure -l docker -l devops

# Debugging lesson
nmem m add "CORS preflight fails silently in fetch. Check Network tab for OPTIONS request, not just the main request." \
  -t "CORS Debugging: Check OPTIONS Request" \
  -i 0.8 --unit-type learning -l debugging -l web

Links

• Documentation
• Nowledge Mem
• Discord Community