Fix agent tool support, fix many other bugs

backend/src/inference.py

@@ -57,7 +57,7 @@ def format_openai(messages):
             formatted.append({
                 'role': 'tool',
                 'tool_call_id': msg.content.id,
-                'content': json.dumps(msg.content.output),
+                'content': json.dumps(msg.content.result),
             })
         else:
             fmsg = {'role': msg.role, 'content': msg.content}
@@ -337,6 +337,9 @@ async def openrouter_chat_completion(app, messages, model, tools=[], opts={}, ag
             'model': model,
             'messages': format_openai(messages),
             'stream': True,
+            # Options
+            'timeout_ms': 45_000,
+            'retries': 3,
         }
 
         # Pass tools in OpenAI-compatible format if provided
@@ -344,63 +347,64 @@ async def openrouter_chat_completion(app, messages, model, tools=[], opts={}, ag
         if formatted_tools:
             kwargs['tools'] = formatted_tools
 
-        try:
-            stream_response = await app.openrouter_client.chat.send_async(**kwargs)
+        stream_response = await app.openrouter_client.chat.send_async(**kwargs)
 
-            # OpenAI format streams tool calls in parts
-            # We could, but I won't assume any tool call is complete until the message is over
-            tool_calls = []
+        # OpenAI format streams tool calls in parts
+        # We could, but I won't assume any tool call is complete until the message is over
+        tool_calls = []
 
-            async for event in stream_response:
-                if agent_run and agent_run.cancelled:
-                    print(f"Cancelling openrouter inference for run {agent_run.id}")
-                    # TODO: Need to cancell?
-                    break
-                    
-                if event.choices and len(event.choices) > 0:
-                    delta = event.choices[0].delta
+        async for event in stream_response:
+            if agent_run and agent_run.cancelled:
+                print(f"Cancelling openrouter inference for run {agent_run.id}")
+                # TODO: Need to cancell?
+                break
+                
+            if event.choices and len(event.choices) > 0:
+                delta = event.choices[0].delta
 
-                    if delta.content:
-                        emit('output', delta.content)
-                    if delta.reasoning:
-                        emit('reasoning', delta.reasoning)
-                    if delta.tool_calls:
-                        print("[openrouter] tool call:" + repr(delta.tool_calls)) 
-                        for tc in delta.tool_calls:
-                            if tc.index >= len(tool_calls):
-                                tool_calls.append(ToolCall(tool_name="", arguments="", id=""))
-                            tool_call = tool_calls[tc.index]
+                if delta.content:
+                    emit('output', delta.content)
+                if delta.reasoning:
+                    emit('reasoning', delta.reasoning)
+                if delta.tool_calls:
+                    print("[openrouter] tool call:" + repr(delta.tool_calls)) 
+                    for tc in delta.tool_calls:
+                        if tc.index >= len(tool_calls):
+                            tool_calls.append(ToolCall(tool_name="", arguments="", id=""))
+                        tool_call = tool_calls[tc.index]
 
-                            if tc.id:
-                                tool_call.id += tc.id
+                        if tc.id:
+                            tool_call.id += tc.id
 
-                            if tc.function:
-                                if tc.function.name:
-                                    tool_call.tool_name += tc.function.name
-                                if tc.function.arguments:
-                                    tool_call.arguments += tc.function.arguments
-                else:
-                    print("[openrouter] no event choices: " + repr(event))
+                        if tc.function:
+                            if tc.function.name:
+                                tool_call.tool_name += tc.function.name
+                            if tc.function.arguments:
+                                tool_call.arguments += tc.function.arguments
+            else:
+                print("[openrouter] no event choices: " + repr(event))
 
-            for tc in tool_calls:
-                try:
-                    arguments = json.loads(tc.arguments)
-                    tc.arguments = arguments
-                    emit('tool-call', tc)
-                except Exception:
-                    print("[openrouter] Bad tool call arguments for " + repr(tc))
-                    print(traceback.format_exc())
-        except asyncio.CancelledError:
-            print(f"[openrouter] task cancelled for run {agent_run.id}")
-            raise
-        except Exception:
-            print(f"[openrouter] exception during inference")
-            print(traceback.format_exc())
+        for tc in tool_calls:
+            try:
+                arguments = json.loads(tc.arguments)
+                tc.arguments = arguments
+                emit('tool-call', tc)
+            except Exception:
+                print("[openrouter] Bad tool call arguments for " + repr(tc))
+                print(traceback.format_exc())
 
         emit('close')
         print("END OPENROUTER INFERENCE")
 
+    def handle_done(task):
+        try:
+            task.result()
+        except Exception as e:
+            print(f"[openrouter] Worker error: {e}")
+            print(traceback.format_exc())
+
     task = asyncio.create_task(worker())
+    task.add_done_callback(handle_done)
     agent_run.task = task