Interface ChatCompletionParams

interface ChatCompletionParams {
    appSource?: string;
    attribution?: WorkloadAttributionInput;
    autoExecuteTools?: boolean;
    billingMode?: BillingMode;
    frequency_penalty?: number;
    max_tokens?: number;
    maxToolRounds?: number;
    messages: ChatMessage[];
    min_p?: number;
    model: string;
    onToolCall?: (toolCall: ToolCall) => Promise<string>;
    onToolProgress?: (
        toolCall: ToolCall,
        progress: ToolExecutionProgress,
    ) => void;
    presence_penalty?: number;
    repetition_penalty?: number;
    response_format?: ChatResponseFormat;
    safeContentFilter?: boolean;
    sogni_tool_execution?: boolean;
    sogni_tools?: SogniToolsMode;
    stop?: string | string[];
    stream?: boolean;
    taskProfile?: "general" | "coding" | "reasoning";
    temperature?: number;
    think?: boolean;
    tokenType?: "sogni" | "spark";
    tool_choice?: ToolChoice;
    tools?: ToolDefinition[];
    top_k?: number;
    top_p?: number;
}

Properties

appSource? attribution? autoExecuteTools? billingMode? frequency_penalty? max_tokens? maxToolRounds? messages min_p? model onToolCall? onToolProgress? presence_penalty? repetition_penalty? response_format? safeContentFilter? sogni_tool_execution? sogni_tools? stop? stream? taskProfile? temperature? think? tokenType? tool_choice? tools? top_k? top_p?

Properties

OptionalappSource

appSource?: string

Optional source label for this request. Defaults to the client appSource when configured.

Optionalattribution

attribution?: WorkloadAttributionInput

Optional workload attribution overriding this client's defaults.

OptionalautoExecuteTools

autoExecuteTools?: boolean

Automatically execute Sogni tool calls (image/video/music generation) when the model requests them. The SDK handles the full multi-round tool calling loop: send completion → execute tools → feed results back → repeat until done.

Only supported with stream: false. For streaming, use chat.tools.executeAll() manually in your own loop.

OptionalbillingMode

billingMode?: BillingMode

Billing mode for this request. 'subscription' bills the job against the active subscription entitlement, 'tokens' forces token billing, and 'auto' (the server default when omitted) lets the server decide. Explicit 'subscription' requests the entitlement cannot cover fail with one of the SUBSCRIPTION_ERROR_CODES exported from the package root.

Optionalfrequency_penalty

frequency_penalty?: number

Optionalmax_tokens

max_tokens?: number

OptionalmaxToolRounds

maxToolRounds?: number

Maximum number of tool calling rounds when autoExecuteTools is enabled. Prevents infinite loops. Default: 5.

messages

messages: ChatMessage[]

Optionalmin_p

min_p?: number

model

model: string

OptionalonToolCall

onToolCall?: (toolCall: ToolCall) => Promise<string>

Handler for non-Sogni tool calls when autoExecuteTools is enabled. Called for any tool call whose name does NOT start with sogni_. Must return a string (the tool result content).

OptionalonToolProgress

onToolProgress?: (toolCall: ToolCall, progress: ToolExecutionProgress) => void

Progress callback for Sogni tool execution when autoExecuteTools is enabled. Fires as each tool call progresses through creation, queuing, processing, and completion.

Optionalpresence_penalty

presence_penalty?: number

Optionalrepetition_penalty

repetition_penalty?: number

Optionalresponse_format

response_format?: ChatResponseFormat

Constrain output structure (OpenAI-compatible). Most useful on tool-call rounds where the model must emit a specific argument shape — eliminates JSON drift on quantized models. Forwarded to the worker unchanged.

OptionalsafeContentFilter

safeContentFilter?: boolean

Requested content-filter policy. The server remains authoritative.

Optionalsogni_tool_execution

sogni_tool_execution?: boolean

When sogni_tools is enabled, ask the Sogni API to execute requested Sogni tools server-side before returning the chat response. Independent of the SDK's client-side autoExecuteTools loop.

Optionalsogni_tools

sogni_tools?: SogniToolsMode

Ask the Sogni API to inject server-side media-generation tool families. true or 'creative-tools' injects the hosted creative media/planning tool family. 'rich' is accepted as a legacy alias for 'creative-tools'. 'creative-agent' adds hosted workflow-control and asset-manifest tools.

Optionalstop

stop?: string | string[]

Optionalstream

stream?: boolean

OptionaltaskProfile

taskProfile?: "general" | "coding" | "reasoning"

Hint for server-side preset selection.

Optionaltemperature

temperature?: number

Optionalthink

think?: boolean

Control thinking/reasoning mode for supported models (e.g. Qwen3/3.5). When false, sends chat_template_kwargs: { enable_thinking: false } so the model skips its internal reasoning step. When true, explicitly enables thinking. When omitted, server defaults apply.

OptionaltokenType

tokenType?: "sogni" | "spark"

Token type to use for billing. Defaults to 'sogni'.

Optionaltool_choice

tool_choice?: ToolChoice

Controls which (if any) tool is called by the model.

Optionaltools

tools?: ToolDefinition[]

Tool definitions for function calling.

Optionaltop_k

top_k?: number

Optionaltop_p

top_p?: number

Settings

Member Visibility

On This Page

Properties