feat: add option to omit tool_choice param for Azure/LiteLLM compatibility

Adds openAiOmitToolChoice setting to the OpenAI provider that allows users
to optionally omit the tool_choice parameter from API requests. This
resolves 400 errors when using Azure-hosted models (like GPT-5) through
LiteLLM proxies that do not support the tool_choice parameter.

Closes: #10701

packages/types/src/provider-settings.ts

@@ -250,6 +250,7 @@ const openAiSchema = baseProviderSettingsSchema.extend({
 	openAiStreamingEnabled: z.boolean().optional(),
 	openAiHostHeader: z.string().optional(), // Keep temporarily for backward compatibility during migration.
 	openAiHeaders: z.record(z.string(), z.string()).optional(),
+	openAiOmitToolChoice: z.boolean().optional(), // Omit tool_choice param for Azure/LiteLLM compatibility.
 })
 
 const ollamaSchema = baseProviderSettingsSchema.extend({

src/api/providers/openai.ts

@@ -161,7 +161,7 @@ export class OpenAiHandler extends BaseProvider implements SingleCompletionHandl
 				...(isGrokXAI ? {} : { stream_options: { include_usage: true } }),
 				...(reasoning && reasoning),
 				...(metadata?.tools && { tools: this.convertToolsForOpenAI(metadata.tools) }),
-				...(metadata?.tool_choice && { tool_choice: metadata.tool_choice }),
+				...(metadata?.tool_choice && !this.options.openAiOmitToolChoice && { tool_choice: metadata.tool_choice }),
 				...(metadata?.toolProtocol === "native" &&
 					metadata.parallelToolCalls === true && {
 						parallel_tool_calls: true,
@@ -231,7 +231,7 @@ export class OpenAiHandler extends BaseProvider implements SingleCompletionHandl
 					? convertToR1Format([{ role: "user", content: systemPrompt }, ...messages])
 					: [systemMessage, ...convertToOpenAiMessages(messages)],
 				...(metadata?.tools && { tools: this.convertToolsForOpenAI(metadata.tools) }),
-				...(metadata?.tool_choice && { tool_choice: metadata.tool_choice }),
+				...(metadata?.tool_choice && !this.options.openAiOmitToolChoice && { tool_choice: metadata.tool_choice }),
 				...(metadata?.toolProtocol === "native" &&
 					metadata.parallelToolCalls === true && {
 						parallel_tool_calls: true,
@@ -358,7 +358,7 @@ export class OpenAiHandler extends BaseProvider implements SingleCompletionHandl
 				reasoning_effort: modelInfo.reasoningEffort as "low" | "medium" | "high" | undefined,
 				temperature: undefined,
 				...(metadata?.tools && { tools: this.convertToolsForOpenAI(metadata.tools) }),
-				...(metadata?.tool_choice && { tool_choice: metadata.tool_choice }),
+				...(metadata?.tool_choice && !this.options.openAiOmitToolChoice && { tool_choice: metadata.tool_choice }),
 				...(metadata?.toolProtocol === "native" &&
 					metadata.parallelToolCalls === true && {
 						parallel_tool_calls: true,
@@ -394,7 +394,7 @@ export class OpenAiHandler extends BaseProvider implements SingleCompletionHandl
 				reasoning_effort: modelInfo.reasoningEffort as "low" | "medium" | "high" | undefined,
 				temperature: undefined,
 				...(metadata?.tools && { tools: this.convertToolsForOpenAI(metadata.tools) }),
-				...(metadata?.tool_choice && { tool_choice: metadata.tool_choice }),
+				...(metadata?.tool_choice && !this.options.openAiOmitToolChoice && { tool_choice: metadata.tool_choice }),
 				...(metadata?.toolProtocol === "native" &&
 					metadata.parallelToolCalls === true && {
 						parallel_tool_calls: true,

webview-ui/src/components/settings/providers/OpenAICompatible.tsx

@@ -173,6 +173,16 @@ export const OpenAICompatible = ({
 				onChange={handleInputChange("openAiUseAzure", noTransform)}>
 				{t("settings:modelInfo.useAzure")}
 			</Checkbox>
+			<div>
+				<Checkbox
+					checked={apiConfiguration?.openAiOmitToolChoice ?? false}
+					onChange={handleInputChange("openAiOmitToolChoice", noTransform)}>
+					{t("settings:modelInfo.omitToolChoice")}
+				</Checkbox>
+				<div className="text-sm text-vscode-descriptionForeground ml-6">
+					{t("settings:modelInfo.omitToolChoiceTips")}
+				</div>
+			</div>
 			<div>
 				<Checkbox
 					checked={azureApiVersionSelected}

webview-ui/src/i18n/locales/en/settings.json

@@ -878,6 +878,8 @@
 		"enableR1Format": "Enable R1 model parameters",
 		"enableR1FormatTips": "Must be enabled when using R1 models such as QWQ to prevent 400 errors",
 		"useAzure": "Use Azure",
+		"omitToolChoice": "Omit tool_choice parameter",
+		"omitToolChoiceTips": "Enable this if your Azure or LiteLLM endpoint doesn't support the tool_choice parameter",
 		"azureApiVersion": "Set Azure API version",
 		"gemini": {
 			"freeRequests": "* Free up to {{count}} requests per minute. After that, billing depends on prompt size.",