AI Settings · Ryokan

Ryokan lets you control how your AI responds. From creativity and response length to managing long conversations, various settings help you customize your experience.

Most options can be adjusted with just a few clicks. For additional control, Ryokan also offers a Power User mode.

Adjusting Response Behavior

Open Settings → API & Model. Under Response behavior, you'll find the main text generation settings.

Creativity (Temperature): Controls how predictable or varied the AI's responses are. Lower values generally produce more consistent results, while higher values allow more variation.
Response Length (Max Tokens): Sets the maximum number of tokens an AI response may use. A higher limit allows longer responses but doesn't force the model to produce them.
Repetition Penalty: Reduces repeated words and phrases. Values that are too high may make the writing sound unnatural.

Ryokan provides predefined options for these settings, so you don't have to work with numerical values directly.

Sampling and Additional Parameters

Under Sampling and other parameters, you'll find additional ways to control text generation.

Word Variety (Top-P) controls how broad the selection of possible next tokens is. Option Limit (Top-K) restricts the selection to a specific number of likely tokens.

The Quality Filter (Min-P) can exclude very unlikely options. Reduce Repetition (Frequency Penalty) makes frequently used words less likely during further generation.

Not every AI provider supports all parameters. Unsupported settings may be ignored by the provider or cause a request to fail.

Reasoning and Thinking Budgets

For supported models, you can adjust how much effort the AI spends on internal processing before responding.

Reasoning offers different effort levels depending on the selected model. Choosing Auto uses the provider's default behavior.

Some models also support a Thinking budget, which limits the number of thinking tokens. A higher budget may allow more complex processing but can increase response times.

With Anthropic and Gemini, thinking and the final response share the output token limit. Increasing the thinking budget does not automatically increase that limit.

Context Window and Memory

Open Settings → Memory to control how Ryokan handles longer conversations.

The Context Window determines how much conversation history the AI can consider at once. Ryokan attempts to detect the available context size of your selected model automatically.

You can choose between three context strategies:

Economy: Uses less context and may reduce API costs.
Balanced: Balances conversation continuity with efficient context usage.
Maximum Context: Uses as much of the available context window as practical.

Alternatively, enable Manual limit to define your own maximum context size.

Conversation Summary

When Conversation summary is enabled, Ryokan can automatically summarize older sections of long conversations.

This helps preserve important information from earlier in the story, even when the original messages no longer fit within the model's context window.

Under Summary connection, you can also select a separate AI connection for generating summaries. For example, you could use a less expensive model for this task.

Creating and Using Presets

Under Settings → Presets, you can create reusable AI configurations.

A preset can contain generation settings and additional prompt instructions. These include System instructions and Final instructions, which are added after the conversation history.

You can save, edit, apply, import, and export presets as JSON files.

Presets are managed independently of your AI connections. Applying a preset transfers the relevant settings to your active connection.

Power User Mode

Open Settings → Advanced to enable Power User mode.

This replaces simple preset buttons with more precise sliders.

Additional options are also available, including:

Service tier: Choose between Auto, Standard, and Flex where supported by the provider.
Additional API Parameters: Supply provider-specific settings using a JSON object.

These options are primarily intended for users who want more precise control over their models.

Changing Settings During a Chat

You don't have to leave your current story to adjust AI settings.

Open the three-dot menu in the chat header. From there, you can change your active AI connection, selected model, and generation parameters.

Changes made through this chat settings panel are saved immediately for subsequent requests.

Tip: The default settings are usually a good starting point. If you want to experiment, try adjusting creativity and response length first before changing several sampling parameters at once.