Skip to main content

​
What Is MEMANTO?

MEMANTO is a universal memory layer for AI agents built on Moorcheh’s no-indexing semantic database.

​
Problem Statement

Large language models (LLMs) have a fundamental limitation: they forget. Each conversation starts fresh with no context from previous interactions. For agents operating in the real world, this is a critical problem:
  • Customer support agents can’t remember customer preferences
  • Project managers forget decisions made in past meetings
  • Research assistants lose context between sessions
  • Learning systems don’t carry knowledge forward

​
Solution: MEMANTO

MEMANTO gives AI agents persistent, semantic memory that:
  • Persists across sessions - Information survives conversation boundaries
  • Is semantically queryable - Recall by meaning, not just keywords
  • Operates at scale - Manage thousands of memories efficiently
  • Costs nothing at idle - Serverless architecture
  • Retrieves instantly - No indexing delays

​
Core Architecture

Your Agent
    ↓ (Store/Recall)
MEMANTO API
    ↓ (Semantic Indexing)
Moorcheh.ai (No-Indexing Semantic Database)

​
Key Concepts

​
1. Agents

An agent is a persistent identity with its own memory namespace: 
Agent = Customer Support Bot
  └── Memory Namespace: memanto_agent_customer-support
       ├── Fact: "Alice prefers email"
       ├── Preference: "Concise responses"
       ├── Event: "Resolved ticket #123"
       └── ... more memories
Each agent maintains independent memories isolated from other agents.

​
2. Sessions

A session is a 6-hour window where an agent is active: 
Session A (Day 1, 8am-2pm)
  ├── Store: Customer preferences
  └── Recall: Use preferences in responses

          (6 hours)

Session B (Day 1, 3pm-9pm)
  ├── Store: New customer issues
  ├── Recall: All memories from Session A still available
  └── (All memories persist across sessions)

          (Gap)

Session C (Day 2, 9am-3pm)
  ├── Recall: Memories from both Session A and B
  └── Continue with full context

​
3. Memories

A memory is a semantic unit of information: 
{
  "id": "mem_abc123",
  "content": "User prefers email communication",
  "type": "preference",
  "confidence": 0.98,
  "created_at": "2025-03-26T09:00:00Z"
}
Memory types categorize information:
  • fact - Objective information
  • preference - Likes/dislikes
  • decision - Choices made
  • commitment - Promises made
  • … and 9 more types
Instead of exact matching, MEMANTO searches by meaning: 
Query: "How should we contact the user?"
Results:
  ✓ "User prefers email communication"
  ✓ "Prefers to be contacted during business hours"
  ✗ "User's email is alice@example.com" (not relevant to method)

​
Why Moorcheh?

MEMANTO uses Moorcheh.ai instead of traditional vector databases:
FeatureTraditional VDBMoorcheh
Write-to-SearchMinutes (indexing)Instant
AccuracyApproximate (ANN)Exact
Idle Costs$$$$ (always running)$0
ComputationHeavy (indexing)Efficient
This means:
  • Memories are available immediately after storage
  • Exact, predictable search results
  • Zero costs when not in use
  • 80% cost savings vs. traditional systems

​
Use Cases

​
Customer Support

Agent remembers customer preferences, past issues, and communication style: 
User: "How should I reach out to Alice?"
Agent recalls: "Email, during PST business hours, prefers concise"

​
Project Management

Agent tracks decisions, commits, and project context: 
Query: "What did we decide about the database?"
Result: "Chose PostgreSQL to support JSON queries"

​
Research Assistant

Agent learns and retains research context across sessions: 
Session 1: Collect papers on ML
Session 2: Analyze papers → recall from Session 1
Session 3: Synthesize findings → recall from Sessions 1 & 2

​
Multi-Agent Systems

Agents coordinate using shared or independent memory: 
Agent A (Sales): Remembers customer budget, needs
Agent B (Support): Remembers customer technical issues, preferences
Shared Context: Company policies, product features

​
How It Works

​
1. Store Memory

Agent: "Store: User prefers dark mode"
    ↓
MEMANTO: Semantically indexes "dark mode preference"
    ↓
Moorcheh: Stores with instant availability

​
2. Recall Memory

Agent: "Recall: What's the UI preference?"
    ↓
MEMANTO: Semantic search for "UI preference"
    ↓
Moorcheh: Returns "User prefers dark mode"

​
3. Get AI Answer

Agent: "Answer: Based on memories, what theme should we use?"
    ↓
MEMANTO: Recalls relevant memories
    ↓
MEMANTO: Uses LLM to generate grounded answer
    ↓
Result: "User prefers dark mode, so use dark theme by default"

​
Comparison

​
Without MEMANTO

Session 1: Agent learns customer prefers email
Session 2: Agent forgets → treats as new customer
Session 3: Painful onboarding again

​
With MEMANTO

Session 1: Agent learns and stores "email preference"
Session 2: Agent recalls preference → personalized service
Session 3: Agent has full customer history

​
Getting Started

​
Quick Steps

  1. Install: pip install memanto
  2. Configure: memanto (set API key)
  3. Create agent: memanto agent create my-bot
  4. Activate: memanto agent activate my-bot
  5. Store memory: memanto remember "My first memory" --type fact
  6. Recall: memanto recall "What did I remember?"

​
Learn More

MEMANTO makes AI agents smarter by giving them the ability to remember. Build the next generation of context-aware AI systems!