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

# Memory Engine

> Configure how your character remembers information about users and conversations — memory types, decay, importance thresholds, and scoring weights.

# Memory Engine

This is where you control **how your character remembers things**. Without memory, your character forgets every conversation the moment it ends. With memory turned on, your character can remember what users said to it days, weeks, or even months ago.

**How to get here:**

> Go to your character's dashboard → left sidebar → click **Universal** → click **Memory Engine**

The full path is: **Universal → Memory Engine**

This section has **two sub-pages**:

1. **Configuration** — settings for how memory works (the "rules")
2. **Memories** — the actual list of things your character currently remembers

***

## Configuration

**Full path: Universal → Memory Engine → Configuration**

The Configuration page has **four tabs** across the top: **General**, **Decay**, **Importance**, and **Scoring**.

***

### General Tab

**Full path: Universal → Memory Engine → Configuration → General**

This is the master control panel for the memory system.

**System Settings**

| Setting                            | What it does                                                                                                                                                                                                                                                 | Default |
| ---------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ------- |
| **Enable Memory System** (toggle)  | The on/off switch for the entire memory system. Turn it off and the character remembers nothing between conversations. Turn it on and memory is active.                                                                                                      | On      |
| **Memory Context Size** (1–20)     | How many memories the character "looks at" before writing each reply. Think of it like how many sticky notes it checks before answering. Higher = checks more notes = smarter but slower.                                                                    | 10      |
| **Relevance Threshold** (0.01–1.0) | How closely a memory has to match the current conversation topic before the character uses it. Think of it like a confidence score. Low = the character uses more memories even if they're not closely related. High = only very relevant memories get used. | 0.10    |
| **Cleanup Threshold** (0.01–0.5)   | Memories with an importance score below this number get automatically deleted during cleanup. Low = keep almost everything. High = only keep very important memories.                                                                                        | varies  |

<Tip>
  If responses feel slow, try lowering **Memory Context Size** to 5 or 7. The character will check fewer memories per reply, which speeds things up.
</Tip>

***

### Decay Tab

**Full path: Universal → Memory Engine → Configuration → Decay**

This controls how fast memories "fade away" over time. Just like how humans forget things they haven't thought about in a while, your character's memories also fade if they aren't reinforced.

**Rule:** Higher number = fades faster. Lower number = lasts longer.

Each type of memory has its own decay rate because different types of memories are naturally expected to last different amounts of time:

| Memory Type    | Default Decay Rate | How long it typically lasts |
| -------------- | ------------------ | --------------------------- |
| **Short-term** | 0.1000             | Hours to a day or two       |
| **Working**    | 0.0500             | A few days                  |
| **Episodic**   | 0.0100             | A few weeks                 |
| **Semantic**   | 0.0010             | Several months              |
| **Long-term**  | 0.0001             | Years                       |

Think of it like this:

* **Short-term** = like remembering you just had a sandwich
* **Working** = like remembering a task you're working on this week
* **Episodic** = like remembering a specific event or conversation from last month
* **Semantic** = like remembering a fact you learned years ago
* **Long-term** = like remembering your own name

***

### Importance Tab

**Full path: Universal → Memory Engine → Configuration → Importance**

This controls the **minimum importance score** a memory must have to survive cleanup. Memories below the threshold for their type are removed or ignored.

Think of it like a filter: "Only keep memories that are important enough."

| Memory Type    | Default Threshold | Timeframe     |
| -------------- | ----------------- | ------------- |
| **Short-term** | 0.30              | Hours to days |
| **Working**    | 0.40              | Days          |
| **Episodic**   | 0.50              | Weeks         |
| **Semantic**   | 0.60              | Months        |
| **Long-term**  | 0.70              | Years         |

* **Raise a threshold** = the character becomes more selective — only high-importance memories survive for that type.
* **Lower a threshold** = the character keeps more of everything, even minor details.

***

### Scoring Tab

**Full path: Universal → Memory Engine → Configuration → Scoring**

This controls **how the system scores memories** when deciding which ones to pull up during a conversation. The three weights must always add up to exactly **1.0** (100%).

Think of it like deciding what's most important when searching your own memory:

* Do you prioritize memories related to the current topic?
* Do you prioritize the most important memories overall?
* Do you prioritize the most recent memories?

| Weight                | Default | What it prioritizes                                           |
| --------------------- | ------- | ------------------------------------------------------------- |
| **Similarity Weight** | 0.25    | Memories that are topically close to the current conversation |
| **Importance Weight** | 0.50    | Memories that were rated as more important when stored        |
| **Recency Weight**    | 0.25    | Memories that were created or last reinforced recently        |

At the bottom of this tab you'll see a **Total Weight** display that shows the live sum of all three. It must read **1.00** — the page will warn you if it doesn't.

**Practical adjustments:**

* Want the character to focus on recent events? → Increase **Recency Weight**.
* Want the character to stay on-topic? → Increase **Similarity Weight**.
* Want all memories treated equally? → Lower **Importance Weight** and raise the others.

***

## Memories

**Full path: Universal → Memory Engine → Memories**

This is the list of everything your character currently remembers. You can search it, filter it, and manage individual memories.

### Stored Memories

| Control                | What it does                                                                           |
| ---------------------- | -------------------------------------------------------------------------------------- |
| **Search bar**         | Type any word or phrase to search through all stored memories.                         |
| **All Types dropdown** | Filter the list to only show one type of memory (short-term, working, episodic, etc.). |
| **Total count**        | Shows how many memories currently exist total.                                         |

You can click on any individual memory to inspect it, and you can delete individual memories you no longer want.

### Learning Queue

Below the memory list, there is a **Learning Queue** section. This is a list of facts the character picked up from conversations that it thinks might be worth remembering permanently.

Each item in the queue shows:

* The fact it learned (e.g., "The user said their name is Alex")
* A category tag (e.g., `user fact`, `world fact`)
* A score out of 5 indicating how likely it is to be a useful memory

For each item, you can:

* Click **Approve** → the fact becomes a permanent memory
* Click **Reject** → the fact is discarded
* Use the **Search bar** to find specific items
* Click **Refresh** to reload the queue with any new items

<Info>
  The Learning Queue is where your character "learns" from real conversations. Check it periodically to approve important facts and reject irrelevant ones. This keeps your character's memory accurate and useful.
</Info>

***

## Danger Zone

At the very bottom of the **Memories** page, there is a section called **Danger Zone** with two permanent actions:

| Action                   | What it does                                                                                                                                   |
| ------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------- |
| **Wipe All Memories**    | Permanently deletes EVERY memory the character has accumulated. The character starts completely fresh with no memory of any past conversation. |
| **Clear Learning Queue** | Removes all items from the Learning Queue without approving any of them.                                                                       |

<Warning>
  Both of these actions are permanent and cannot be undone. Once memories are wiped, they are gone forever. The character will not remember any past conversations.
</Warning>
