Fix agent tool support, fix many other bugs

backend/src/agents/expert.py

@@ -24,11 +24,14 @@ class ExpertAgent(Agent):
     async def run(self, run):
         opts = {}
 
-        using_tools = False
+        # Is this roon using *Agent's* tools, as opposed to run tools
+        using_agent_tools = False
         if 'tools' not in run.input:
+            # Agent tools will fully replace run tools if none are present,
+            # otherwise they won't be used
             if self.tools:
                 run.input['tools'] = self.tools
-                using_tools = True
+                using_agent_tools = True
 
         if self.config.get("temperature") is not None:
             opts["temperature"] = self.config["temperature"]
@@ -52,12 +55,11 @@ class ExpertAgent(Agent):
             output_stream = stream
         run.on('output', on_output)
         run.on('inference-end', lambda event: inference_end.set())
-        run.on('tool-call', lambda tc: tool_calls.append(tc))
+        run.on('tool-call', lambda tc: print(f"[expert] tool call: {tc}") or tool_calls.append(tc))
 
         while True:
             # Reset data
             inference_end.clear()
-            tool_calls = []
             output_stream = None
 
             # Call inference, passing the run to get events out
@@ -73,8 +75,10 @@ class ExpertAgent(Agent):
             await inference_end.wait()
 
             messages.append(inference.LlamaMessage('assistant', output_stream and await output_stream.content_wait() or ""))
+            print(f"[expert] collected tool calls: {len(tool_calls)}, using? {using_agent_tools}")
 
-            if using_tools and tool_calls:
+            if using_agent_tools and tool_calls:
+                print("Will execute tool calls")
                 for tc in tool_calls:
                     tool = next(t for t in self.tools if t.name == tc.tool_name)
                     if not tool:
@@ -82,7 +86,11 @@ class ExpertAgent(Agent):
                         continue
                     tc.result = await tool.execute_async(**tc.arguments)
                     messages.append(inference.LlamaMessage('tool', tc))
+                tool_calls = []
+                # Buffer time to let openrouter rest?
+                await asyncio.sleep(2)
             else:
+                print("No tool calls to execute, finish")
                 break
         
         run.close()