Mnemex

# Mnemex Use Cases Based on the repository documentation and architecture, here are the appropriate use cases for Mnemex: ## 1. **Personal AI Assistant Memory** **Scenario**: You regularly chat with Claude about various topics - Remember your **preferences** (coding style, communication preferences, dietary restrictions) - Recall **past decisions** and their reasoning - Track **ongoing projects** and their status - Remember **personal context** (family names, pet names, important dates) **Example**: "I prefer tabs over spaces" gets saved once, reinforced over time, and Claude remembers it months later without you repeating it. --- ## 2. **Software Development Assistant** **Scenario**: Using Claude for coding across multiple projects - Remember **architecture decisions** and rationale - Track **bugs you've encountered** and solutions - Recall **library preferences** and why you chose them - Remember **API patterns** you've established - Track **tech debt** and future refactoring notes **Example**: Claude remembers you prefer React hooks over class components, your ESLint config preferences, and that one weird TypeScript issue you solved last month. --- ## 3. **Context Switching for Developers** **Scenario**: Jumping between multiple codebases/projects - **Aggressive forgetting** for ephemeral context - Quick recall of **project-specific conventions** - Remember **which commands to run** for each project - Track **environment setup quirks** **Example**: Set shorter decay (1-day half-life) so context from Project A fades quickly when you switch to Project B, preventing confusion. --- ## 4. **Research & Learning** **Scenario**: Using Claude to learn new topics or conduct research - Build a **knowledge graph** of concepts and their relationships - Remember **key insights** from papers/articles - Track **questions to explore** later - Link **related concepts** across domains - **Spaced repetition** naturally surfaces concepts that need review **Example**: Learning Rust - Claude remembers the ownership rules you struggled with and brings them up when relevant, strengthening that memory through use. --- ## 5. **Writing & Content Creation** **Scenario**: Working on long-form content with Claude - Remember **style guidelines** and tone preferences - Track **character details** for fiction writing - Recall **research findings** relevant to your topic - Remember **audience preferences** and feedback - Track **ideas for future articles** **Example**: Writing a blog series - Claude remembers your target audience, writing style, and callbacks to previous posts without you re-explaining each time. --- ## 6. **Personal Knowledge Management (PKM)** **Scenario**: Building a second brain with Obsidian integration - **Auto-generate Obsidian notes** from conversations - Link memories to your **existing note structure** - Automatic **tagging and entity extraction** - Build connections between **conversation insights** and permanent notes - Search across both **ephemeral and permanent** knowledge **Example**: Your Obsidian vault becomes enriched with conversation insights that auto-promote to permanent notes when they prove valuable over time. --- ## 7. **Preference-Heavy Applications** **Scenario**: Any domain where user preferences matter a lot - **Design preferences** (color schemes, layouts) - **Workflow preferences** (automation preferences, tool choices) - **Communication style** (formal vs casual, emoji usage) - **Accessibility needs** (screen reader usage, keyboard shortcuts) **Example**: Claude remembers you're colorblind and always suggests colorblind-friendly palettes without asking. --- ## 8. **Long-Term Projects & Planning** **Scenario**: Multi-month projects with Claude as a collaborator - Track **project goals** and evolution - Remember **stakeholder feedback** - Recall **past iterations** and why they changed - Monitor **progress milestones** - Link related **sub-projects** and dependencies **Example**: Building a SaaS product over 6 months - Claude remembers your MVP scope, feature requests you've deferred, and technical constraints. --- ## 9. **Team Knowledge Sharing** **Scenario**: Shared memory store for team AI interactions - Document **team conventions** and decisions - Build **institutional knowledge** graph - Remember **common problems** and solutions - Track **who knows what** (entity linking) - Create searchable **decision log** **Example**: Team members can query why certain architectural decisions were made, with memories strengthening as multiple people reference them. --- ## 10. **Domain-Specific Expertise** **Scenario**: Using Claude in specialized domains - **Medical/Healthcare**: Remember patient interaction patterns (anonymized) - **Legal**: Track case precedents and reasoning - **Education**: Remember student learning patterns - **Finance**: Recall market analysis insights - **Scientific Research**: Build knowledge graphs of experiments and findings **Example**: A researcher remembers which experiments failed and why, preventing repeated mistakes. --- ## 11. **Adaptive AI Behavior** **Scenario**: You want Claude's behavior to adapt over time - **Natural spaced repetition** - memories in danger of forgetting surface naturally - **Cross-context detection** - memories used in multiple domains get stronger - **Automatic importance weighting** - frequently-used memories survive - **Graceful forgetting** - ephemeral context naturally fades **Example**: Claude stops asking about your Python version after the 5th conversation where it's mentioned and used. --- ## When NOT to Use Mnemex ❌ **High-security secrets** - Use proper secret management (see `docs/security.md`) ❌ **Regulated data (PHI, PII)** - Compliance concerns unless properly configured ❌ **Real-time high-throughput** - Designed for assistant conversations, not APIs ❌ **Exact recall requirements** - Temporal decay means memories can be forgotten ❌ **Multi-user production systems** - Single-user design (as of v0.5.3) --- ## Configuration for Different Use Cases From `docs/configuration.md` and `src/mnemex/config.py:1`: ### For Development (fast context switching) ```json { "decay_half_life_hours": 24, "forget_threshold": 0.1, "review_lower_bound": 0.2 } ``` ### For Research/Archival (long retention) ```json { "decay_half_life_hours": 168, "forget_threshold": 0.03, "promote_threshold": 0.5 } ``` ### For Personal Assistant (balanced) ```json { "decay_half_life_hours": 72, "forget_threshold": 0.05, "promote_threshold": 0.65 } ``` --- ## Bottom Line Mnemex is best for **individual knowledge workers** who want their AI assistant to remember context across conversations, with memory dynamics that feel natural rather than robotic. It's particularly powerful when combined with **Obsidian** for building a hybrid ephemeral/permanent knowledge base. The temporal decay ensures Claude doesn't get confused by outdated context, while the reinforcement mechanics ensure important information naturally persists.