AxModelConfig

type AxModelConfig = object;

Defined in: https://github.com/ax-llm/ax/blob/05ff5bd88d050f7ba85a3fcc6eb0ed2975ad7d51/src/ax/ai/types.ts#L75

Configuration options for AI model behavior.

These settings control how the model generates responses. They can be set as defaults when creating an AI instance, or overridden per-request.

Example

const config: AxModelConfig = {
  maxTokens: 2000,
  temperature: 0.7,
  topP: 0.9
};

Properties

endSequences?

optional endSequences: string[];

Defined in: https://github.com/ax-llm/ax/blob/05ff5bd88d050f7ba85a3fcc6eb0ed2975ad7d51/src/ax/ai/types.ts#L171

Similar to stopSequences, but the sequence IS included in the output.

Example

['</answer>'] to include closing tag in output

frequencyPenalty?

optional frequencyPenalty: number;

Defined in: https://github.com/ax-llm/ax/blob/05ff5bd88d050f7ba85a3fcc6eb0ed2975ad7d51/src/ax/ai/types.ts#L154

Penalizes tokens based on how frequently they’ve appeared. Range: -2.0 to 2.0.

Unlike presencePenalty, this scales with frequency: tokens that appear many times get penalized more. Useful for preventing the model from repeating the same phrases verbatim.

Example

0.5 to discourage word/phrase repetition

maxTokens?

optional maxTokens: number;

Defined in: https://github.com/ax-llm/ax/blob/05ff5bd88d050f7ba85a3fcc6eb0ed2975ad7d51/src/ax/ai/types.ts#L89

Maximum number of tokens to generate in the response.

Token estimation guide:

• ~750 tokens ≈ 1 page of English text
• ~100 tokens ≈ 75 words
• ~4 characters ≈ 1 token (English)

Set higher for long-form content (articles, code), lower for concise responses (classifications, short answers).

Example

500 for short responses, 2000 for detailed explanations, 4000+ for long-form content

n?

optional n: number;

Defined in: https://github.com/ax-llm/ax/blob/05ff5bd88d050f7ba85a3fcc6eb0ed2975ad7d51/src/ax/ai/types.ts#L189

Number of completions to generate for each prompt.

Generates multiple independent responses. Useful with result pickers to select the best response. Increases cost proportionally.

Example

3 to generate three alternatives and pick the best

presencePenalty?

optional presencePenalty: number;

Defined in: https://github.com/ax-llm/ax/blob/05ff5bd88d050f7ba85a3fcc6eb0ed2975ad7d51/src/ax/ai/types.ts#L142

Penalizes tokens that have already appeared in the output. Range: -2.0 to 2.0.

Positive values reduce repetition by penalizing tokens that have appeared at all, regardless of frequency. Useful for encouraging diverse vocabulary.

• 0 - No penalty (default)
• 0.5-1.0 - Mild penalty, reduces obvious repetition
• 1.5-2.0 - Strong penalty, may hurt coherence

Example

0.6 to reduce repetitive phrasing

stopSequences?

optional stopSequences: string[];

Defined in: https://github.com/ax-llm/ax/blob/05ff5bd88d050f7ba85a3fcc6eb0ed2975ad7d51/src/ax/ai/types.ts#L164

Sequences that will stop generation when encountered.

The model stops generating as soon as any stop sequence is produced. The stop sequence itself is NOT included in the output.

Example

['\\n\\n', 'END', '---'] to stop at double newlines or markers

stream?

optional stream: boolean;

Defined in: https://github.com/ax-llm/ax/blob/05ff5bd88d050f7ba85a3fcc6eb0ed2975ad7d51/src/ax/ai/types.ts#L179

Enable streaming responses for real-time output.

When true, the response is returned as a stream of chunks, allowing you to display partial results as they’re generated.

temperature?

optional temperature: number;

Defined in: https://github.com/ax-llm/ax/blob/05ff5bd88d050f7ba85a3fcc6eb0ed2975ad7d51/src/ax/ai/types.ts#L104

Controls randomness in generation. Range: 0 to 2.

Use case guide:

• 0 - Deterministic, always picks most likely token. Best for factual Q&A, classification, code generation where consistency matters.
• 0.3-0.5 - Low creativity. Good for structured outputs, summaries.
• 0.7 - Balanced (default for most models). Good for general conversation.
• 1.0 - High creativity. Good for brainstorming, creative writing.
• 1.5-2.0 - Very high randomness. Often produces incoherent output.

Default

Varies by provider, typically 0.7-1.0

topK?

optional topK: number;

Defined in: https://github.com/ax-llm/ax/blob/05ff5bd88d050f7ba85a3fcc6eb0ed2975ad7d51/src/ax/ai/types.ts#L127

Only consider the top K most likely tokens at each step.

Lower values (e.g., 10-40) make output more focused. Not supported by all providers (OpenAI doesn’t support this; Anthropic, Google do).

Example

40 for focused output, 100 for more variety

topP?

optional topP: number;

Defined in: https://github.com/ax-llm/ax/blob/05ff5bd88d050f7ba85a3fcc6eb0ed2975ad7d51/src/ax/ai/types.ts#L117

Nucleus sampling: only consider tokens with cumulative probability >= topP. Range: 0 to 1.

Lower values make output more focused and deterministic. Alternative to temperature for controlling randomness.

Recommendation: Adjust either temperature OR topP, not both.

Example

0.1 for focused output, 0.9 for diverse output