fix: Ai Chat: do not send tools if empty + respond even if tool fails (#5692)

* do not send tools if empty

* respond to user request even if tool call fails

* remove test line

* add missing await

frontend/src/lib/components/copilot/chat/core.ts

@@ -487,6 +487,31 @@ async function callTool(
 	}
 }
 
+async function processToolCall(
+	toolCall: ChatCompletionMessageToolCall,
+	messages: ChatCompletionMessageParam[],
+	lang: ScriptLang | 'bunnative'
+) {
+	try {
+		const args = JSON.parse(toolCall.function.arguments)
+		let result = ''
+		try {
+			result = await callTool(toolCall.function.name, args, lang, get(workspaceStore) ?? '')
+		} catch (err) {
+			console.error(err)
+			result =
+				'Error while calling tool, MUST tell the user to check the browser console for more details, and then respond as much as possible to the original request'
+		}
+		messages.push({
+			role: 'tool',
+			tool_call_id: toolCall.id,
+			content: result
+		})
+	} catch (err) {
+		console.error(err)
+	}
+}
+
 export async function chatRequest(
 	messages: ChatCompletionMessageParam[],
 	abortController: AbortController,
@@ -554,23 +579,7 @@ export async function chatRequest(
 						tool_calls: toolCalls
 					})
 					for (const toolCall of toolCalls) {
-						try {
-							const args = JSON.parse(toolCall.function.arguments)
-							const result = await callTool(
-								toolCall.function.name,
-								args,
-								lang,
-								get(workspaceStore) ?? ''
-							)
-							messages.push({
-								role: 'tool',
-								tool_call_id: toolCall.id,
-								content: result
-							})
-						} catch (err) {
-							console.error(err)
-							throw new Error('Error while calling tool')
-						}
+						await processToolCall(toolCall, messages, lang)
 					}
 				} else {
 					break

frontend/src/lib/components/copilot/lib.ts

@@ -385,7 +385,7 @@ function getProviderAndCompletionConfig<K extends boolean>({
 				: {
 						model: modelProvider.model,
 						temperature: 0,
-						tools
+						...(tools && tools.length > 0 ? { tools } : {})
 					}),
 			max_tokens: getModelMaxTokens(modelProvider.model),
 			messages: processedMessages,