Connection presets
A connection preset saves the AI service and model Museris uses to write replies. You can make more than one. For example, keep one dependable preset for everyday chats and another for a more creative story. Choose the one you want to use from Connection.
Create and test a connection
- Open Connection and select Create Preset.
- Give the preset a recognizable name.
- Choose OpenAI Compatible or Museris Cloud.
- For an OpenAI-compatible service, enter its base URL and API key if it needs one.
- Choose a generation model from the model list.
- Select Test Connection, then save the preset.
Museris loads the model list after it has a valid URL and, if needed, an API key. If the list is empty, check the URL, key, and models available from your service.
Museris Cloud
Sign in to Museris Hub before selecting Museris Cloud. Its connection details are supplied automatically; select one of the available models and save the preset.
OpenAI-compatible services
Enter the service's API base URL, including http:// or https://. Museris works with OpenAI-compatible APIs and asks the service for models at its /models address. Leave the API key blank only if the service does not use one.
Select a preset
Click a preset in the Connection list to make it active. Roleplay and instruct chats use the active preset when they are opened. You can keep presets for different services or model settings and switch between them without retyping their details.
Tune generation
Leave a setting blank to use the model or service default. Some services ignore settings they do not support.
Temperature
Controls how freely the AI makes things up. Lower values give safer, more predictable replies. Higher values give more variety and surprises. If a character starts to drift, lower it a little. If replies feel repetitive, raise it a little.
Reasoning effort
Controls how much thinking a supported model does before it replies. Provider Default is the safest choice when you are unsure. A service may change or ignore levels it does not support.
Max response tokens
Sets the longest reply Museris can ask for. A larger number allows longer replies, but may take more time and use more of your provider's allowance. The model may still stop sooner.
Seed
A seed can make generation more repeatable when the provider supports it. The same seed and settings can still produce different results if the provider, model, prompt, or server state changes.
Top P, Top K, and Min P
These settings limit the words a model considers as it writes.
- Top P keeps the most likely choices until their combined probability reaches a threshold. Lower values usually make output more focused; higher values allow more variety.
- Top K limits the model to a fixed number of its most likely choices. Smaller values are more constrained.
- Min P discards choices that are too unlikely relative to the most likely choice. It can reduce strange outliers while preserving some variety.
Start with temperature alone. Using several sampling controls at once can make a preset harder to predict.
Presence and frequency penalties
Both settings reduce repetition. Presence Penalty makes the model less likely to return to topics already mentioned. Frequency Penalty grows when words or phrases repeat. It can reduce repeated phrases in a long chat. Large penalties can make a character sound unnatural, so change them a little at a time.
Advanced provider options
Provider Options (JSON) sends extra settings to an OpenAI-compatible service. Use it for service features that do not have a Museris control. It must be valid JSON. Reasoning settings placed here override the Reasoning Effort setting above.
Only add options you understand from your provider's documentation. Invalid or unsupported options can cause generation requests to fail.