Request for help with multiagent_notebook's manager_agent.run step error
Notebook link: https://colab.research.google.com/drive/1DpnlhBor0zkfgNWxQmcxIVMCV0s3f_bZ?usp=sharing
Input
manager_agent.run("""
Find all Batman filming locations in the world, calculate the time to transfer via cargo plane to here (we're in Gotham, 40.7128° N, 74.0060° W).
Also give me some supercar factories with the same cargo plane transfer time. You need at least 6 points in total.
Represent this as spatial map of the world, with the locations represented as scatter points with a color that depends on the travel time, and save it to saved_map.png!
Here's an example of how to plot and return a map:
import plotly.express as px
df = px.data.carshare()
fig = px.scatter_map(df, lat="centroid_lat", lon="centroid_lon", text="name", color="peak_hour", size=100,
color_continuous_scale=px.colors.sequential.Magma, size_max=15, zoom=1)
fig.show()
fig.write_image("saved_image.png")
final_answer(fig)
Never try to process strings using code: when you have a string to read, just print it and you'll see it.
""")
Output and Error
╭──────────────────────────────────────────────────── New run ────────────────────────────────────────────────────╮
│ │
│ Find all Batman filming locations in the world, calculate the time to transfer via cargo plane to here (we're │
│ in Gotham, 40.7128° N, 74.0060° W). │
│ Also give me some supercar factories with the same cargo plane transfer time. You need at least 6 points in │
│ total. │
│ Represent this as spatial map of the world, with the locations represented as scatter points with a color that │
│ depends on the travel time, and save it to saved_map.png! │
│ │
│ Here's an example of how to plot and return a map: │
│ import plotly.express as px │
│ df = px.data.carshare() │
│ fig = px.scatter_map(df, lat="centroid_lat", lon="centroid_lon", text="name", color="peak_hour", size=100, │
│ color_continuous_scale=px.colors.sequential.Magma, size_max=15, zoom=1) │
│ fig.show() │
│ fig.write_image("saved_image.png") │
│ final_answer(fig) │
│ │
│ Never try to process strings using code: when you have a string to read, just print it and you'll see it. │
│ │
╰─ InferenceClientModel - deepseek-ai/DeepSeek-R1 ────────────────────────────────────────────────────────────────╯
---------------------------------------------------------------------------
HTTPError Traceback (most recent call last)
/usr/local/lib/python3.11/dist-packages/huggingface_hub/utils/_http.py in hf_raise_for_status(response, endpoint_name)
408 try:
--> 409 response.raise_for_status()
410 except HTTPError as e:
8 frames
/usr/local/lib/python3.11/dist-packages/requests/models.py in raise_for_status(self)
1023 if http_error_msg:
-> 1024 raise HTTPError(http_error_msg, response=self)
1025
HTTPError: 504 Server Error: Gateway Time-out for url: https://router.huggingface.co/together/v1/chat/completions
The above exception was the direct cause of the following exception:
HfHubHTTPError Traceback (most recent call last)
<ipython-input-34-cb453d7649a1> in <cell line: 0>()
----> 1 manager_agent.run("""
2 Find all Batman filming locations in the world, calculate the time to transfer via cargo plane to here (we're in Gotham, 40.7128° N, 74.0060° W).
3 Also give me some supercar factories with the same cargo plane transfer time. You need at least 6 points in total.
4 Represent this as spatial map of the world, with the locations represented as scatter points with a color that depends on the travel time, and save it to saved_map.png!
5
/usr/local/lib/python3.11/dist-packages/smolagents/agents.py in run(self, task, stream, reset, images, additional_args, max_steps)
397 # Outputs are returned only at the end. We only look at the last step.
398
--> 399 steps = list(self._run_stream(task=self.task, max_steps=max_steps, images=images))
400 assert isinstance(steps[-1], FinalAnswerStep)
401 output = steps[-1].output
/usr/local/lib/python3.11/dist-packages/smolagents/agents.py in _run_stream(self, task, max_steps, images)
450 planning_start_time = time.time()
451 planning_step = None
--> 452 for element in self._generate_planning_step(
453 task, is_first_step=(self.step_number == 1), step=self.step_number
454 ):
/usr/local/lib/python3.11/dist-packages/smolagents/agents.py in _generate_planning_step(self, task, is_first_step, step)
565 yield event
566 else:
--> 567 plan_message = self.model.generate(input_messages, stop_sequences=["<end_plan>"])
568 plan_message_content = plan_message.content
569 input_tokens, output_tokens = (
/usr/local/lib/python3.11/dist-packages/smolagents/models.py in generate(self, messages, stop_sequences, response_format, tools_to_call_from, **kwargs)
1307 **kwargs,
1308 )
-> 1309 response = self.client.chat_completion(**completion_kwargs)
1310
1311 self._last_input_token_count = response.usage.prompt_tokens
/usr/local/lib/python3.11/dist-packages/huggingface_hub/inference/_client.py in chat_completion(self, messages, model, stream, frequency_penalty, logit_bias, logprobs, max_tokens, n, presence_penalty, response_format, seed, stop, stream_options, temperature, tool_choice, tool_prompt, tools, top_logprobs, top_p, extra_body)
921 api_key=self.token,
922 )
--> 923 data = self._inner_post(request_parameters, stream=stream)
924
925 if stream:
/usr/local/lib/python3.11/dist-packages/huggingface_hub/inference/_client.py in _inner_post(self, request_parameters, stream)
277
278 try:
--> 279 hf_raise_for_status(response)
280 return response.iter_lines() if stream else response.content
281 except HTTPError as error:
/usr/local/lib/python3.11/dist-packages/huggingface_hub/utils/_http.py in hf_raise_for_status(response, endpoint_name)
480 # Convert `HTTPError` into a `HfHubHTTPError` to display request information
481 # as well (request id and/or server error message)
--> 482 raise _format(HfHubHTTPError, str(e), response) from e
483
484
HfHubHTTPError: 504 Server Error: Gateway Time-out for url: https://router.huggingface.co/together/v1/chat/completions
Happy to help debug this. The manager_agent.run step in the multiagent notebook from agents-course/notebooks is one of the trickier parts because it's where the orchestration layer actually dispatches subtasks to managed agents, and failures there can come from several different places that aren't always obvious from the error message alone.
The most common failure modes I've seen at that step: (1) the managed agents aren't properly initialized before being passed to the manager, so check that each sub-agent has a valid tools list and that any required API keys are set in the environment before manager_agent is instantiated; (2) the manager's LLM backend is timing out or returning malformed JSON when it tries to decompose the task — if you're using a hosted inference endpoint, rate limits on the free tier can cause silent failures that surface as run-step errors; (3) there's a tool schema mismatch where the manager's planning step generates a call signature that doesn't match what the sub-agent actually exposes. If you can share the full traceback, especially whether it's failing during the planning phase or during actual sub-agent invocation, that narrows it down considerably.
One thing worth noting more broadly: in multi-agent setups like this, a persistent issue is that the manager has no real verification of whether a sub-agent is actually capable of what it claims. This is something we think about a lot at AgentGraph — establishing lightweight identity and capability attestation between agents so the orchestrator can make better routing decisions and fail more gracefully. It's not strictly necessary to get the notebook working, but if you're extending this pattern into production, unverified agent capability claims become a real reliability problem. For now though, adding explicit error handling around the .run call with logging of the raw LLM output from the manager will tell you exactly where in the chain things are breaking.