Memori

Architecture

Memori Cloud is a managed memory platform for AI applications. Connect your LLM client, set attribution, and Memori handles the rest — storage, augmentation, knowledge graph construction, and recall.

System Overview

"Memori Cloud System Overview"

Core Components

Your Application

Your code and existing LLM client. It sends requests through the Memori SDK and receives model responses as usual.

Memori SDK

The integration layer between your app and Memori Cloud. It provides LLM wrappers, attribution, and the Recall API.

Memori Cloud

The managed backend that processes captured conversations and powers storage, augmentation, and recall services.

Managed Storage

Stores conversations, sessions, and facts for each attribution scope.

Advanced Augmentation

Processes raw conversation data into structured memory through fact extraction, embeddings, and knowledge graph construction.

Recall Engine

Surfaces the right memories at the right time — semantic search over stored memory, intelligent ranking and decay, and seamless injection of relevant context into every LLM call so your AI stays contextually aware.

Configuration

Setting up Memori requires only your API key and attribution:

from memori import Memori
from openai import OpenAI

# Set MEMORI_API_KEY as an environment variable
# export MEMORI_API_KEY="your-memori-api-key"

client = OpenAI()
mem = Memori().llm.register(client)
mem.attribution(entity_id="user_123", process_id="my_agent")

Data Flow

  1. Conversation Capture — Every LLM call through the wrapped client is captured and sent to Memori Cloud. Your app gets the response immediately.

  2. Attribution Tracking — Attribution links every conversation to a specific entity and process so memories are properly scoped and indexed.

  3. Augmentation — After a conversation completes, Memori Cloud processes it asynchronously — extracts facts, generates embeddings, and builds knowledge graph triples.

  4. Recall — On the next LLM call, Memori embeds the query, performs vector search across the entity's stored facts, and injects the most relevant memories into the context.