Function createStream

src/api/providers/base-openai-compatible-provider.ts:70–111

(
		systemPrompt: string,
		messages: Anthropic.Messages.MessageParam[],
		metadata?: ApiHandlerCreateMessageMetadata,
		requestOptions?: OpenAI.RequestOptions,
	)

	}

	protected createStream(
		systemPrompt: string,
		messages: Anthropic.Messages.MessageParam[],
		metadata?: ApiHandlerCreateMessageMetadata,
		requestOptions?: OpenAI.RequestOptions,
	) {
		const { id: model, info } = this.getModel()

		// Centralized cap: clamp to 20% of the context window (unless provider-specific exceptions apply)
		const max_tokens =
			getModelMaxOutputTokens({
				modelId: model,
				model: info,
				settings: this.options,
				format: "openai",
			}) ?? undefined

		const temperature = this.options.modelTemperature ?? info.defaultTemperature ?? this.defaultTemperature

		const params: OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming = {
			model,
			max_tokens,
			temperature,
			messages: [{ role: "system", content: systemPrompt }, ...convertToOpenAiMessages(messages)],
			stream: true,
			stream_options: { include_usage: true },
			tools: this.convertToolsForOpenAI(metadata?.tools),
			tool_choice: metadata?.tool_choice,
			parallel_tool_calls: metadata?.parallelToolCalls ?? true,
		}

		// Add thinking parameter if reasoning is enabled and model supports it
		if (this.options.enableReasoningEffort && info.supportsReasoningBinary) {
			;(params as any).thinking = { type: "enabled" }
		}

		try {
			return this.client.chat.completions.create(params, requestOptions)
		} catch (error) {
			throw handleOpenAIError(error, this.providerName)
		}
	}

	override async *createMessage(
		systemPrompt: string,