Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.memorylake.ai/llms.txt

Use this file to discover all available pages before exploring further.

​
Writing Effective Memories

​
Be Specific and Concrete

Bad: 
This project is about customer data.
Good: 
This project analyzes Q1 2024 customer survey responses from 1,500 participants.

Focus areas:
- Net Promoter Score trends
- Product satisfaction ratings
- Feature request themes

Survey methodology: Online survey, random sampling, 95% confidence interval

​
Provide Context

Help readers (human and AI) understand:
  • Why this project exists
  • What problem it solves
  • Who the audience is
  • How it relates to other projects

​
Use Clear Structure

Organize content with sections: 
PURPOSE
Brief description of project goals

CONTEXT
Background information

KEY INFORMATION
Important facts, figures, decisions

GUIDELINES
How to approach the content

RELATED
Links to related projects or external resources

​
Naming Conventions

​
Descriptive Names

Make names self-explanatory:
  • ✅ “Budget Constraints Q1 2024”
  • ✅ “Customer Interview - Key Insights”
  • ❌ “Notes 1”
  • ❌ “Misc”

​
Consistent Patterns

Choose a pattern and stick to it: Topic-Based:
  • “Research Methodology”
  • “Research Questions”
  • “Research Timeline”
Category Prefixes:
  • “[Context] Project History”
  • “[Guidelines] Analysis Approach”
  • “[Reference] Related Studies”

​
Content Guidelines

​
Length

Ideal: 100-500 words per memory Too Short (< 50 words): 
Project about sales data.
Not enough context for AI or collaborators. Too Long (> 1000 words): Hard to scan, better split into multiple focused memories. Just Right: 
This project analyzes sales data from our Q4 2023 campaign across three regions: North America, Europe, and Asia Pacific.

Objectives:
1. Compare regional performance against targets
2. Identify top-performing products
3. Understand seasonal trends
4. Recommend Q1 2024 strategies

Data sources: CRM exports, Google Analytics, manual sales reports

When analyzing, focus on year-over-year growth rather than absolute numbers, as we expanded team size significantly in 2023.

Related projects: "Q3 2023 Sales Analysis", "2024 Sales Planning"

​
Update Regularly

Memories should stay current:
  • Update when information changes
  • Add new insights as project progresses
  • Remove outdated information
  • Archive historical context separately

​
Use Cases by Memory Type

​
Project Guidelines

Provide instructions for working with the project: 
ANALYSIS GUIDELINES

When reviewing these documents:
1. Prioritize data from the last 6 months
2. Cross-reference customer feedback with NPS scores
3. Focus on actionable insights over descriptive statistics
4. Consider budget constraints ($50K for Q1)

Red flags to watch for:
- Sample sizes under 100
- Survey response rates under 30%
- Conflicting data between sources

​
Background Context

Explain project history: 
PROJECT HISTORY

This project began in September 2023 after customer complaints about checkout process increased 40%.

Initial hypothesis: Payment gateway issues causing abandonment

Investigation revealed:
- Gateway was fine
- UI confusion was main culprit
- Mobile experience particularly poor

This analysis supports the redesign project (see "Checkout Redesign 2024")

​
Key Decisions

Document important choices: 
TECHNICAL DECISIONS

Why we chose PostgreSQL over MySQL:
- Better JSON support for flexible schemas
- Superior full-text search
- Team expertise
- Cost-effective at our scale

Trade-offs accepted:
- Slightly more complex setup
- Fewer hosting options

Decision date: January 15, 2024
Decision makers: Engineering team, CTO approval

​
Constraints and Limitations

Note boundaries: 
PROJECT CONSTRAINTS

Budget: $75,000 (hard limit)
Timeline: Must complete by March 31, 2024
Team: 3 developers, 1 designer (part-time)
Scope: MVP only - advanced features for Phase 2

Known Limitations:
- Data only available from 2022 forward
- API rate limited to 1000 requests/hour
- No access to competitor pricing data

​
Integration with MCP Servers

When memories will be read by AI models:

​
Be Explicit

AI doesn’t have implicit context: Vague: 
Focus on the important stuff.
Explicit: 
When analyzing documents:
1. Prioritize files in /reports/final/ over /drafts/
2. Pay special attention to customer quotes (marked with >>)
3. Ignore internal process notes (in /admin/ folders)

​
Provide Examples

Help AI understand expectations: 
RESPONSE STYLE

When answering questions about this project, use this style:

Example good response:
"According to the Q1 report (page 12), customer satisfaction increased 15% after implementing the new checkout flow. This aligns with the NPS improvement documented in the March survey."

Example poor response:
"Things got better."

Always cite specific documents and page numbers when available.

​
Common Mistakes to Avoid

❌ Too vague: “This is an important project” ✅ Specific: “This project determines Q2 budget allocation based on Q1 results” ❌ Kitchen sink: One memory with everything ✅ Focused: Multiple memories, each with clear purpose ❌ Stale: Last updated 2 years ago ✅ Current: Updated as project evolves ❌ Internal jargon: “Use the TPS reports for the Q-analysis” ✅ Clear: “Use the Transaction Processing System reports for quarterly analysis” ❌ No structure: Wall of text ✅ Organized: Clear sections with line breaks

​
Next Steps

Memories Overview

Complete guide to creating and managing memories

MCP Servers

Provide AI access to your memories