Skip to main content
Before running a prompt, you need to select which LLM processes it and configure how that model behaves. The Editor’s model settings panel gives you complete control over provider selection, generation parameters, and response formatting. Managing an LLM's settings in Adaline

​
Select an LLM

Adaline’s Editor displays all supported LLMs based on the providers you have configured in your workspace settings. Open the model selector to browse and choose your preferred model: Selecting an LLM in Adaline Each model is displayed in the format provider::model_name. This prefix helps you distinguish between provider accounts when you have multiple keys configured for the same provider — for example, OpenAI-dev::gpt-4o and OpenAI-prod::gpt-4o.
To add a new AI provider, navigate to your workspace settings. See Configure AI Provider for setup instructions.

​
Configure generation settings

Click the settings icon next to the model selector to open the configuration panel: Changing LLM settings in Adaline's Editor The available parameters depend on the selected model. Common settings include:
ParameterDescriptionTypical range
TemperatureControls randomness in responses. Lower values produce more deterministic outputs; higher values increase creativity.0 – 2
Max TokensSets the maximum number of tokens the model can generate in a single response.Model-dependent
Top PControls diversity via nucleus sampling. The model considers tokens whose cumulative probability reaches this threshold.0 – 1
Frequency PenaltyReduces repetition by penalizing tokens that have already appeared frequently.-2 – 2
Presence PenaltyEncourages the model to introduce new topics by penalizing tokens that have appeared at all.-2 – 2
The interface automatically shows only the parameters that are relevant to the model you have selected. Different providers and models support different subsets of these settings.

​
Configure response format

You can control the structure of the model’s output by configuring the response format. Click on Response format in the settings panel: Configure an LLM's response format Choose from the following format options:
FormatOutputWhen to use
textFree-form text (default)General-purpose prompts — summarization, Q&A, creative writing, chat.
json_objectValid JSON in an auto-determined schemaQuick prototyping when you need structured output but don’t need a fixed schema. Your prompt must contain the word “json” (case insensitive).
json_schemaJSON adhering to a strict schema you defineProduction workflows — API integrations, structured data extraction, pipelines that depend on a consistent response shape.

​
Define a JSON schema

When using json_schema, provide a schema definition in the response format editor. Here is an example: 
{
  "strict": true,
  "name": "users_data",
  "description": "A schema for creating user data.",
  "schema": {
    "type": "object",
    "properties": {
      "user_city": {
        "type": "string",
        "description": "The city where the user resides."
      },
      "user_name": {
        "type": "string",
        "description": "The full name of the user."
      },
      "user_age": {
        "type": "number",
        "description": "The age of the user in years."
      }
    },
    "required": ["user_name", "user_age", "user_city"],
    "additionalProperties": false
  }
}
Key requirements for the schema:
  • The "strict": true field is mandatory.
  • The name field is mandatory and must use underscores (e.g., users_data) or camelCase (e.g., usersData) — no spaces or special characters.
  • Define the structure inside the "schema" field using standard JSON Schema syntax.
The schema you provide must adhere to the OpenAI JSON schema structure. This format is used across all supported providers.
The image below shows a prompt configured with a JSON schema and the structured response from the model: Schema-based responses in Adaline

​
Next steps

Use Roles in Prompts

Structure prompts with role-based messages and multi-shot techniques.

Use Tools in Prompts

Enable function calling and tool use for your selected model.