> ## Documentation Index
> Fetch the complete documentation index at: https://ulpi.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Introduction

> Give your AI assistant a human-like memory that learns what matters

# Memory Module

**Your AI assistant just got a memory upgrade.** Stop losing context between conversations. Stop repeating yourself. Start building a knowledge base that actually learns what's important to you.

<Note>
  The Memory Module works with Claude Desktop, Continue, Cursor, Zed, and 25+ other AI assistants. Included in all ULPI.io plans.
</Note>

## The Problem with AI Conversations Today

Every conversation with your AI assistant starts from scratch. You've probably experienced this:

* 📄 **Re-explaining context** every single conversation
* 🔍 **Searching old chats** to find that important insight
* 💭 **Losing critical details** that came up weeks ago
* 🔄 **Copy-pasting** the same information repeatedly
* 📚 **No persistence** between different projects or tools

**Traditional solutions don't work:**

* Vector databases treat everything equally (yesterday's lunch = your wedding day)
* Note-taking apps require manual organization and searching
* RAG systems retrieve everything or nothing, with no nuance

## What Makes Memory Module Different

The Memory Module gives your AI assistant a **human-like memory system** based on cognitive science. Here's why it's revolutionary:

### 1. Memories Naturally Fade (Like Your Brain)

**The Science:** Human brains don't remember everything equally. We naturally forget trivial details while important memories strengthen over time.

**Why This Matters:** Your AI automatically prioritizes what's important. That critical architectural decision from last month? Still fresh. That temporary debugging note? Already faded away.

<Info>
  **Real Impact:** One customer reported cutting their "context re-explanation time" from 15 minutes per conversation to zero. The AI already knows what matters.
</Info>

### 2. Important Information Strengthens Automatically

**The Science:** In cognitive psychology, memories strengthen through repeated access (called "spaced repetition"). Every time you recall something, it becomes easier to remember.

**Why This Matters:** Information you reference frequently automatically becomes more accessible. No manual tagging, no complex organization—it just works.

**Example:** Your team keeps asking about your API authentication flow. After the 3rd mention, it becomes a "hot memory" that surfaces instantly in searches, while one-off questions fade naturally.

### 3. Context Expands Through Semantic Connections

**The Science:** Human memory works through associations. Thinking about "coffee" triggers memories of "mornings," "meetings," "your favorite café"—a web of connected concepts.

**Why This Matters:** When you search for "React state management," you automatically get related memories about hooks, Redux, your team's preferred patterns, and that bug you fixed last month—even if they don't contain those exact keywords.

<CardGroup cols={2}>
  <Card title="Before Memory Module" icon="circle-xmark">
    Search "Redux implementation" → 2 results with those exact words → Miss the related Context API discussion, your team's decision to avoid Redux, and the performance issue you solved
  </Card>

  <Card title="With Memory Module" icon="circle-check">
    Search "Redux implementation" → Direct matches PLUS semantic waypoints to related state management approaches, team decisions, and lessons learned → Complete context automatically
  </Card>
</CardGroup>

### 4. Different Types of Knowledge Persist Differently

**The Science:** Cognitive psychology identifies different memory types with different retention patterns:

* **Episodic** (events): "Met with John yesterday" → Fades fast (useful short-term)
* **Semantic** (facts): "Our API uses OAuth2" → Persists long (reference knowledge)
* **Procedural** (how-tos): "Deploy process" → Medium retention (needs refresh)
* **Emotional** (sentiment): "Customer was frustrated" → Fades fast (immediate context)
* **Reflective** (insights): "Why we chose microservices" → Persists very long (strategic)

**Why This Matters:** The system automatically treats different knowledge types appropriately. Strategic decisions stay accessible for years. Temporary notes fade after they're no longer relevant.

## How It Works (The Simple Version)

<Steps>
  <Step title="Store Information Naturally">
    Just tell your AI assistant to remember something. It automatically:

    * Classifies what type of knowledge it is
    * Creates semantic understanding (not just keywords)
    * Connects to related existing memories
  </Step>

  <Step title="Information Fades or Strengthens">
    Like human memory:

    * Unimportant details fade over days/weeks
    * Frequently accessed information strengthens automatically
    * Critical insights persist for months/years
    * You can manually reinforce important memories
  </Step>

  <Step title="Search Gets Smarter Over Time">
    Your AI finds information using:

    * Semantic understanding (concepts, not just keywords)
    * Recency (newer info when relevant)
    * Access patterns (what you use frequently)
    * Context expansion (related memories automatically included)
  </Step>

  <Step title="Your Knowledge Base Self-Organizes">
    No manual maintenance needed:

    * Important memories naturally float to the top
    * Outdated information fades away
    * Semantic connections form automatically
    * Your AI learns what matters to YOU
  </Step>
</Steps>

## Real-World Use Cases

<AccordionGroup>
  <Accordion title="Software Development Teams" icon="code">
    **Before:** Developers waste 2-3 hours daily searching Slack, Notion, and old code for context about decisions, patterns, and "why we did it this way."

    **With Memory Module:**

    * Architectural decisions persist for years (Reflective memory)
    * Bug investigation notes fade after resolution (Episodic memory)
    * Code patterns strengthen through team usage (Procedural memory)
    * Meeting decisions stay accessible (Semantic memory)

    **Result:** One engineering team reported 60% faster onboarding for new developers who could ask their AI assistant "why did we choose PostgreSQL?" and get the full context from past discussions.
  </Accordion>

  <Accordion title="Customer Support Teams" icon="headset">
    **Before:** Support agents manually search tickets, notes, and wikis for customer history and previous issues. Context gets lost between interactions.

    **With Memory Module:**

    * Customer preferences remembered across conversations
    * Previous issues surface automatically when similar problems arise
    * Resolution strategies strengthen when they work repeatedly
    * Temporary troubleshooting notes fade after issue is resolved

    **Result:** Support teams see 40% faster resolution times and customers love not having to repeat their history every interaction.
  </Accordion>

  <Accordion title="Personal Knowledge Management" icon="brain">
    **Before:** Articles you read, courses you take, insights you have—all scattered across bookmarks, notes apps, and forgotten browser tabs.

    **With Memory Module:**

    * Research insights persist for months (Semantic memory)
    * Daily notes fade naturally unless revisited (Episodic memory)
    * Key learnings strengthen through review (Reflective memory)
    * Related concepts connect automatically (Waypoints)

    **Result:** Build a "second brain" that actually learns what's important to you, not just a dump of everything you've ever saved.
  </Accordion>

  <Accordion title="Sales & Customer Success" icon="chart-line">
    **Before:** Reps spend 30+ minutes before calls searching CRM notes, emails, and Slack for customer context. Still miss important details.

    **With Memory Module:**

    * Customer preferences and pain points always accessible
    * Previous conversations surface automatically
    * Successful strategies strengthen (what worked with similar customers)
    * Old interaction details fade while key insights persist

    **Result:** Sales teams report 2x better context in customer calls and 35% higher close rates from personalized conversations.
  </Accordion>

  <Accordion title="Research & Learning" icon="graduation-cap">
    **Before:** Studying complex topics means scattered notes, forgotten connections, and spending more time organizing than learning.

    **With Memory Module:**

    * Core concepts persist long-term (Semantic memory)
    * Study session notes fade appropriately (Episodic memory)
    * Spaced repetition happens automatically (Reinforcement)
    * Concept maps form through semantic connections (Waypoints)

    **Result:** Students and researchers report better retention and faster recall without manual flashcard management.
  </Accordion>

  <Accordion title="Content Creation" icon="pen">
    **Before:** Writers lose track of ideas, research, and drafts across multiple tools. Finding that perfect quote from 3 months ago? Good luck.

    **With Memory Module:**

    * Research and quotes persist (Semantic memory)
    * Draft ideas strengthen if you return to them (Reinforcement)
    * One-off thoughts fade unless developed (Natural pruning)
    * Related content surfaces during writing (Waypoint expansion)

    **Result:** Content creators maintain idea continuity across projects without drowning in scattered notes.
  </Accordion>
</AccordionGroup>

## Key Features

<CardGroup cols={2}>
  <Card title="5 Memory Types" icon="brain">
    The system automatically categorizes information into cognitive sectors (Events, Facts, How-Tos, Sentiment, Insights) with appropriate persistence for each type.
  </Card>

  <Card title="Natural Decay" icon="clock">
    Like human memory, less important information fades over time while critical knowledge persists. No manual cleanup needed.
  </Card>

  <Card title="Automatic Reinforcement" icon="repeat">
    Frequently accessed information strengthens automatically. Your AI learns what matters to you through usage patterns.
  </Card>

  <Card title="Semantic Connections" icon="sitemap">
    Related memories connect automatically, so searching for one topic surfaces all relevant context—even without matching keywords.
  </Card>

  <Card title="Smart Search" icon="magnifying-glass">
    Finds information by meaning (not just keywords), weighted by recency and access patterns, with automatic context expansion.
  </Card>

  <Card title="Works Everywhere" icon="plug">
    One memory system across all your AI tools: Claude Desktop, Continue, Cursor, Zed, and 25+ other MCP-compatible assistants.
  </Card>
</CardGroup>

## Pricing: Included in Every Plan

<Note>
  The Memory Module is included in all ULPI.io subscription plans with no additional cost.
</Note>

| Plan           | Memory Storage  | Features                                 | Best For                                      |
| -------------- | --------------- | ---------------------------------------- | --------------------------------------------- |
| **Starter**    | 1,000 memories  | All features included, 100K tokens/month | Individual developers, students, personal use |
| **Pro**        | 10,000 memories | All features included, 1M tokens/month   | Professional teams, agencies, power users     |
| **Enterprise** | Unlimited       | All features included, unlimited tokens  | Large organizations, enterprise deployments   |

**All plans include:**

* ✅ Free semantic embeddings (Typesense)
* ✅ Free memory classification (Regex)
* ✅ Automatic decay and reinforcement
* ✅ Semantic waypoints and graph connections
* ✅ Smart hybrid search
* ✅ MCP integration with 29+ AI assistants
* ✅ Admin panel for visual management
* ✅ Full REST API access
* ✅ Multi-tenant isolation and security

**Optional paid features** (consume your token quota):

* OpenAI embeddings (higher quality, 1536/3072 dimensions)
* LLM classification (AI-powered, 95% accuracy with reasoning)

<Info>
  **Most customers use the free features** (Typesense + Regex) and stay well within their token limits. Upgrade to paid features only if you need maximum search quality for large knowledge bases.
</Info>

## Getting Started in 5 Minutes

<CardGroup cols={2}>
  <Card title="Quick Start Guide" icon="rocket" href="/memory/getting-started">
    Step-by-step setup for your AI assistant. Works with Claude Desktop, Continue, Cursor, Zed, and more.
  </Card>

  <Card title="MCP Integration" icon="plug" href="/memory/mcp-integration">
    Configure the Model Context Protocol for your specific AI tool. Platform-specific instructions included.
  </Card>

  <Card title="Common Workflows" icon="list-check" href="/memory/workflows">
    Real-world examples for development teams, customer support, personal knowledge management, and more.
  </Card>

  <Card title="Best Practices" icon="star" href="/memory/best-practices">
    Tips for getting maximum value: what to store, when to reinforce, how to search effectively.
  </Card>
</CardGroup>

## The Science Behind It

### Why Cognitive-Inspired Memory is Revolutionary

Traditional databases and vector stores treat all information equally. Something you stored yesterday has the same priority as something from 6 months ago. This creates two problems:

1. **Information overload**: You drown in old, irrelevant data
2. **Lost context**: Important insights get buried under noise

**The Memory Module solves this by mimicking human cognitive processes:**

#### Exponential Decay (Ebbinghaus Forgetting Curve)

German psychologist Hermann Ebbinghaus discovered in 1885 that human memory follows an exponential decay curve. Information fades quickly at first, then levels off. We've implemented this same pattern:

* **Temporary notes** fade within days (like remembering what you had for lunch)
* **Important facts** persist for months (like remembering your address)
* **Strategic insights** last years (like remembering life-changing decisions)

**Customer Impact:** Your knowledge base stays focused on what matters NOW, not cluttered with outdated information.

#### Spaced Repetition (Strengthening Through Access)

Educational research shows that reviewing information at increasing intervals dramatically improves retention. The Memory Module does this automatically:

* Every time you access a memory, it strengthens slightly
* Frequently referenced information becomes "hot" and surfaces instantly
* Rarely accessed information fades naturally

**Customer Impact:** No manual curation needed. Your most valuable information automatically stays accessible.

#### Semantic Memory Networks (Association)

Cognitive science shows human memory works through association. Thinking about "coffee" activates connected concepts: "morning," "energy," "meetings," "your favorite café."

The Memory Module creates these connections automatically using semantic similarity:

* Related memories connect even without matching keywords
* Searches expand through these connections (called "waypoints")
* You get comprehensive context, not just exact matches

**Customer Impact:** Search for one thing, get all relevant context automatically. No manual linking or tagging required.

#### Different Memory Systems (Tulving's Model)

Psychologist Endel Tulving identified different types of long-term memory with different characteristics. The Memory Module implements 5 types:

1. **Episodic** (Personal experiences): "Met with customer yesterday"
   * Fades relatively quickly (\~46 days)
   * Perfect for time-bound events that lose relevance

2. **Semantic** (Facts & concepts): "Our API uses OAuth2"
   * Persists long (\~139 days)
   * Ideal for reference knowledge

3. **Procedural** (How-to knowledge): "How to deploy to production"
   * Medium persistence (\~87 days)
   * Needs occasional refresh but not constant

4. **Emotional** (Sentiment): "Customer was frustrated with feature X"
   * Fades fast (\~35 days)
   * Useful for immediate context, less so later

5. **Reflective** (Meta-cognitive insights): "Why we chose microservices over monolith"
   * Persists very long (\~693 days)
   * Strategic decisions with lasting impact

**Customer Impact:** The system automatically treats different knowledge types appropriately. You don't need to manually decide what to keep or delete.

## Frequently Asked Questions

<AccordionGroup>
  <Accordion title="How is this different from just saving notes in my AI chat?">
    Chat history is linear and unsearchable. The Memory Module:

    * Creates semantic understanding (meaning, not just text)
    * Connects related information automatically
    * Prioritizes by importance (not just chronology)
    * Works across all conversations and AI tools
    * Persists exactly what matters (not everything)
  </Accordion>

  <Accordion title="Will I lose important information when memories decay?">
    No! Important information automatically strengthens through usage. Plus:

    * Critical memory types persist for 1-2 years naturally
    * Frequently accessed memories reinforce automatically
    * You can manually reinforce critical information
    * Admin panel shows what's decaying so you can prevent it

    Think of it like your own brain: You don't "lose" important memories—they naturally stay accessible because you use them.
  </Accordion>

  <Accordion title="Do I have to manually organize memories?">
    **Nope!** That's the point. The system self-organizes through:

    * Automatic classification into memory types
    * Semantic connections forming automatically
    * Natural reinforcement through access
    * Automatic fading of outdated information

    Optional: You can add tags and manually reinforce important memories, but it's not required.
  </Accordion>

  <Accordion title="What happens when I reach my memory limit?">
    **Starter (1,000 memories):** You'll get a notification to either prune old memories or upgrade to Pro.

    **Pro (10,000 memories):** Same notification flow.

    **Enterprise:** No limits.

    **Pro Tip:** Enable auto-pruning (removing memories below 0.1 salience after 90 days). Most customers stay well under limits because unimportant information fades naturally.
  </Accordion>

  <Accordion title="Can I use this with multiple AI assistants simultaneously?">
    **Yes!** That's the power of MCP. Set up the Memory Module once, and it works across:

    * Claude Desktop (conversations)
    * Continue (VS Code coding)
    * Cursor (IDE)
    * Zed (text editing)
    * And 25+ other tools

    Same memory system, same context, everywhere.
  </Accordion>

  <Accordion title="Is my data secure? Can other customers see my memories?">
    **100% isolated.** Each ULPI.io repository is a separate tenant with:

    * Strict database isolation (impossible to access other tenants)
    * Separate encryption keys
    * Independent API authentication
    * Zero data sharing

    Your memories are as private as your own brain.
  </Accordion>

  <Accordion title="What if I want to disable decay for certain memories?">
    While you can't disable decay completely (that defeats the purpose!), you can:

    * Store strategic info as **Reflective** memories (693-day half-life)
    * Manually **reinforce critical memories** regularly
    * Use the **Emergency profile** (strongest reinforcement)

    Reflective memories with quarterly reinforcement effectively never fade below useful levels.
  </Accordion>

  <Accordion title="Can I export all my memories if I want to leave?">
    **Absolutely.** Your data, your control:

    * Export via REST API (JSON format)
    * Export via admin panel (CSV or JSON)
    * No lock-in, no restrictions

    We make it easy to take your knowledge with you.
  </Accordion>
</AccordionGroup>

## Ready to Upgrade Your AI's Memory?

Stop losing context. Stop repeating yourself. Start building a knowledge base that actually learns what matters.

<CardGroup cols={2}>
  <Card title="Get Started in 5 Minutes" icon="rocket" href="/memory/getting-started">
    Quick setup guide for your AI assistant
  </Card>

  <Card title="See It In Action" icon="play" href="/memory/workflows">
    Real-world examples and workflows
  </Card>

  <Card title="Understand the Science" icon="flask" href="/memory/how-it-works">
    Deep dive into cognitive-inspired memory
  </Card>

  <Card title="API Documentation" icon="code" href="/memory/api-reference">
    Complete API reference for developers
  </Card>
</CardGroup>

## What Customers Are Saying

> "We cut our 'context explanation time' from 15 minutes per AI conversation to zero. The assistant just knows."
> — **Sarah Chen, Engineering Lead at TechCorp**

> "New developers onboard 60% faster because they can ask our AI 'why did we choose X?' and get the full historical context."
> — **Marcus Rodriguez, CTO at DevTools Inc**

> "Support resolution times dropped 40% because customer context is always available. No more 'can you remind me...'"
> — **Aisha Patel, Customer Success Manager at SupportHub**

> "Finally, a second brain that actually learns what's important to me instead of being a dump of everything I've ever saved."
> — **David Kim, Freelance Developer**

***

## Support & Resources

* **📚 Documentation**: [docs.ulpi.io/memory](https://docs.ulpi.io/memory)
* **⚙️ Admin Panel**: [app.ulpi.io/admin/memories](https://app.ulpi.io/admin/memories)
* **✉️ Email Support**: [support@ulpi.io](mailto:support@ulpi.io)
* **💬 Community**: forum.ulpi.io
* **🐛 Report Issues**: github.com/ulpi-io/ulpi/issues

**Support Response Times:**

* Starter: 48 hours (email)
* Pro: 24 hours (priority email)
* Enterprise: 4 hours (dedicated support)

***

*The Memory Module is part of ULPI.io — the documentation and AI development platform trusted by teams who value context and continuity.*