# Tokens & Context

Tokens are the pieces of text the AI uses to read your messages and generate replies.

Every word (or part of a word), symbol, and emoji uses tokens.

Understanding **tokens** and the **context window** helps you get longer, more consistent chats.

### What are tokens?

Tokens are the units the AI uses to process text.

Every time you:

* send a message
* receive a reply
* create a character
* write a greeting, scenario, or example dialogue

…you are using tokens.

Token limits depend on your chosen [AI model](/product-guides/premium-features/ai-models.md) and your subscription tier.

### How tokens affect character creation

When creating a character, every field uses tokens:

* personality
* greeting
* scenario
* background and lore
* example dialogues

You can see token usage under each text box.

{% hint style="info" %}
A common target is keeping character setup concise (often around **800–1,100 tokens**). This leaves more room for the conversation.
{% endhint %}

<figure><img src="/files/y8O7o3pJs36JciiGgRCb" alt=""><figcaption></figcaption></figure>

### What is a context window?

The **context window** is the amount of text the AI can “see” at one time. It uses that text to generate the next reply.

Think of it like a desk with limited space:

* your recent messages are on the desk
* the character definition is on the desk
* recent replies are on the desk
* generation instructions and settings are on the desk

When the desk gets full, older messages are removed to make room.

#### Why this matters

If an older message no longer fits, the AI may not use it. That can break continuity in long chats.

Older details stay consistent if they are:

* repeated
* summarized
* stored using memory systems

{% hint style="success" %}
Simple rule: the AI remembers the most recent parts best.
{% endhint %}

### Why older messages get forgotten

AI does not have unlimited live memory during a chat.

As your chat grows, the system must fit all of this into the context window:

* character definition
* your messages
* character replies
* generation settings and instructions

When there is no room left, the oldest messages drop out.

{% hint style="danger" %}
Once a message is outside the current context window, the AI may no longer use it.
{% endhint %}

### Why we don’t always use the model’s maximum context

Modern models can support very large context windows.

Larger context also increases:

* cost
* response time
* compute usage

To keep SpicyChat fast and affordable, we set context limits by tier.

### How the context window fills up (examples)

Your context window is shared by everything the AI needs for a reply, including:

* character definition
* recent messages
* recent replies
* generation instructions and settings

So even if your tier supports 4,096 tokens, not all 4,096 are available for chat history.

#### Example: Free tier (4k token context)

Let’s say:

* character definition = **1,000 tokens**
* average message (user or bot) = **\~150 tokens**

That leaves roughly **3,096 tokens** for recent conversation content. This is before extra formatting and instruction overhead.

At \~150 tokens per message, the AI may only keep around **20 recent messages** visible. This is a rough estimate.

{% hint style="info" %}
Exact numbers vary with message length, character setup size, and generation settings.
{% endhint %}

#### Example: I'm All In tier (16k token context)

With a larger context window, more of the recent chat stays visible.

In many chats, this means dozens more messages stay “in memory”. In short-message chats, it can sometimes be close to 100 previous messages.

{% hint style="info" %}
This is an estimate, not a guarantee. Longer messages reduce how many fit.
{% endhint %}

### Context window by subscription tier

This is the maximum conversation context (the total amount of text the AI can work with at once):

* **Free Users:** Up to **4,096 tokens**
* **Get A Taste Users:** Up to **4,096 tokens**
* **True Supporter Users:** Up to **8,192 tokens**
* **I’m All In Users:** Up to **16,384 tokens**

{% hint style="warning" %}
This total is shared across everything needed for the reply. It is not just your most recent messages.
{% endhint %}

### How this affects long conversations

As a conversation gets longer:

* the AI keeps the most recent messages
* older messages may drop out of the context window
* continuity can weaken over time

{% hint style="warning" %}
If an important detail was mentioned much earlier, repeat it or summarize it.
{% endhint %}

### Reply tokens vs context window

These are not the same thing.

#### 1) Context window

How much total text the AI can see and use at once.

#### 2) Reply tokens

How many tokens the AI can spend generating a single reply.

If reply tokens are too low, responses get shorter or cut off. This can happen even with a large context window.

### Reply token limits by tier

Per-reply tokens depend on your settings and subscription tier:

* **Free and Get A Taste Users:** Up to **180 tokens per reply**
* **True Supporter and I’m All In Users:** Up to **300 tokens per reply**

You can change this in [Generation Settings](/advanced/generation-settings.md).

{% hint style="info" %}
Many users confuse **reply length** with **memory**.

They are related, but they are not the same thing:

* **Reply tokens** = how long a response can be
* **Context window** = how much the AI can “see” at once
  {% endhint %}

### Semantic Memory (subscriber benefit)

Subscriber tiers may also benefit from **Semantic Memory 2.0**.

Semantic memory helps preserve important details from earlier in the chat. It can help even when the original messages no longer fit in the context window.

It improves long-term continuity in long roleplays.

{% hint style="success" %}
Semantic memory helps with continuity when older messages drop out.
{% endhint %}

Learn more in [Semantic Memory 2.0](/product-guides/premium-features/semantic-memory-2.0.md) and [Memory Manager](/product-guides/premium-features/memory-manager.md).

### Getting the most out of your tokens

* Keep character definitions concise and focused
* Avoid long greetings and huge example dialogues unless needed
* Repeat or summarize key details in long chats
* Increase reply tokens in Generation Settings for longer replies
* Use higher tiers for a larger context window and better continuity

### Quick summary

* **Tokens** measure text
* **Reply tokens** affect response length
* **Context window** affects what the AI can use at once
* In long chats, older messages may drop out of active context
* Higher tiers allow larger context windows
* **Semantic memory** helps preserve important details beyond active context

{% hint style="info" %}
If a character forgets something from much earlier, it usually means it no longer fits in the current context window.
{% endhint %}


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.spicychat.ai/advanced/tokens-and-context.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.