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

# MemScene

> Organizing memories into thematic clusters and coherent knowledge

A single memory is rarely useful in isolation. Real intelligence comes from connecting the dots. **MemScene** is the high-level structural unit in EverOS that organizes individual [MemCells](/cloud/concepts/memcell) into coherent, thematic clusters.

## What is a MemScene?

A MemScene represents a specific "context" or "theme" in the user's life or work. It is a dynamic graph of related MemCells.

* **Example**: If you are a developer, you might have a MemScene for "Python Project X," another for "Job Interview Prep," and another for "Personal Hobbies."
* **Function**: When you ask a question about Project X, the system loads the "Project X" MemScene, bringing all relevant history, facts, and files into the Agent's working memory.

## Semantic Consolidation

MemScenes are created and maintained through a background process called **Semantic Consolidation**. This process runs asynchronously (often when the user is not active), similar to how humans consolidate memories during sleep.

<Steps>
  <Step title="Clustering">
    The system analyzes new MemCells using semantic vector embedding. It identifies which existing MemScene the new memory belongs to, or creates a new cluster if the topic is novel.
  </Step>

  <Step title="Synthesis">
    Within a MemScene, the system looks for redundancies or contradictions.

    * *Redundancy*: "I like coffee" and "I drink coffee every day" might be merged.
    * *Contradiction*: "My dog is 3 years old" (old memory) vs "My dog turned 4 today" (new memory). The system updates the facts accordingly.
  </Step>

  <Step title="User Persona Update">
    Key traits extracted from the MemCells are promoted to the global **User Profile**. This is a high-level summary of who the user is (e.g., "Senior Python Developer," "Living in New York," "Prefers concise answers").
  </Step>
</Steps>

## The "Context Window" Problem

LLMs have limited context windows. Even with 1 million tokens, dumping raw logs is inefficient and noisy.

MemScenes solve this by providing a **pre-digested, structured view** of a topic. Instead of reading 500 pages of chat logs to understand "Project X," the Agent reads the "Project X" MemScene, which contains the synthesized narrative, key facts, and current status.

<Info>
  **Dynamic Evolution**: MemScenes are never finished. They grow and change as you interact with the system, ensuring the AI's understanding of a topic evolves just as yours does.
</Info>