Retry for openrouter on max_tokens overage (#2274)

src/backend/src/services/ai/chat/providers/OpenRouterProvider/OpenRouterProvider.ts

@@ -84,7 +84,7 @@ export class OpenRouterProvider implements IChatProvider {
 
         messages = await OpenAIUtil.process_input_messages(messages);
 
-        const completion = await this.#openai.chat.completions.create({
+        const completionParams = {
             messages,
             model: modelIdForParams,
             ...(tools ? { tools } : {}),
@@ -95,7 +95,23 @@ export class OpenRouterProvider implements IChatProvider {
                 stream_options: { include_usage: true },
             } : {}),
             usage: { include: true },
-        } as ChatCompletionCreateParams);
+        } as ChatCompletionCreateParams;
+
+        let completion;
+        try {
+            completion = await this.#openai.chat.completions.create(completionParams);
+        } catch ( e: any ) {
+            // If you overestimate allowed max_tokens on openrouter then it will throw an error.
+            // Since we know the user has enough for the query anyways, we should reexecute the
+            // request without max_tokens.
+            if ( e && e.error && e.error.message && e.error.message.startsWith("This endpoint's maximum context length is ") ) {
+                delete completionParams.max_tokens;
+                completion = await this.#openai.chat.completions.create(completionParams);
+            } else {
+                console.log('Openarouter error: ', e.error.message);
+                throw e;
+            }
+        }
 
         return OpenAIUtil.handle_completion_output({
             usage_calculator: ({ usage }) => {