Questions about DeepResearchAgent

[ww2283]

AG2 version 0.7.4; I did pip install --upgrade git+https://github.com/ag2ai/ag2.git then pip install --upgrade "git+https://github.com/ag2ai/ag2.git#egg=pyautogen[browser-use,crawl4ai,gemini]" as of 2025-02-14.

Question 1: an error inside the loop

When I follow the tutorial at https://docs.ag2.ai/docs/blog/2025-02-13-DeepResearchAgent/index, with gpt-4o, I noticed this kind of error inside the loop

WebSurferAgent (to WebSurferCritic):

***** Suggested tool call (call_AnfPm7tH6srVesBvFj5GAp6l): browser_use *****
Arguments: 
{"task":"Search for information on the financial or economic factors impacting DeepSeek and its influence on stock prices."}
****************************************************************************

--------------------------------------------------------------------------------

>>>>>>>> EXECUTING FUNCTION browser_use...
Call ID: call_AnfPm7tH6srVesBvFj5GAp6l
Input arguments: {'task': 'Search for information on the financial or economic factors impacting DeepSeek and its influence on stock prices.'}
INFO     [agent] 🚀 Starting task: Search for information on the financial or economic factors impacting DeepSeek and its influence on stock prices.
INFO     [agent] 📍 Step 1
INFO     [agent] 👍 Eval: Success - Initial state is an empty page, ready to begin search.
INFO     [agent] 🧠 Memory: Search 0/10 done, start search process for DeepSeek.
INFO     [agent] 🎯 Next goal: Search for information about DeepSeek's financial or economic factors affecting stock prices.
INFO     [agent] 🛠️  Action 1/1: {"search_google":{"query":"DeepSeek financial economic factors impact stock prices"}}
INFO     [controller] 🔍  Searched for "DeepSeek financial economic factors impact stock prices" in Google
INFO     [agent] 📍 Step 2
Task exception was never retrieved
future: <Task finished name='Task-694' coro=<AsyncClient.aclose() done, defined at /Users/wei/opt/anaconda3/envs/autogen/lib/python3.11/site-packages/httpx/_client.py:2024> exception=RuntimeError('Event loop is closed')>
Traceback (most recent call last):
  File "/Users/wei/opt/anaconda3/envs/autogen/lib/python3.11/site-packages/httpx/_client.py", line 2031, in aclose
    await self._transport.aclose()
  File "/Users/wei/opt/anaconda3/envs/autogen/lib/python3.11/site-packages/httpx/_transports/default.py", line 389, in aclose
    await self._pool.aclose()
  File "/Users/wei/opt/anaconda3/envs/autogen/lib/python3.11/site-packages/httpcore/_async/connection_pool.py", line 353, in aclose
    await self._close_connections(closing_connections)
  File "/Users/wei/opt/anaconda3/envs/autogen/lib/python3.11/site-packages/httpcore/_async/connection_pool.py", line 345, in _close_connections
    await connection.aclose()
  File "/Users/wei/opt/anaconda3/envs/autogen/lib/python3.11/site-packages/httpcore/_async/connection.py", line 173, in aclose
    await self._connection.aclose()
  File "/Users/wei/opt/anaconda3/envs/autogen/lib/python3.11/site-packages/httpcore/_async/http11.py", line 258, in aclose
    await self._network_stream.aclose()
  File "/Users/wei/opt/anaconda3/envs/autogen/lib/python3.11/site-packages/httpcore/_backends/anyio.py", line 53, in aclose
    await self._stream.aclose()
  File "/Users/wei/opt/anaconda3/envs/autogen/lib/python3.11/site-packages/anyio/streams/tls.py", line 201, in aclose
    await self.transport_stream.aclose()
  File "/Users/wei/opt/anaconda3/envs/autogen/lib/python3.11/site-packages/anyio/_backends/_asyncio.py", line 1306, in aclose
    self._transport.close()
  File "/Users/wei/opt/anaconda3/envs/autogen/lib/python3.11/asyncio/selector_events.py", line 864, in close
    self._loop.call_soon(self._call_connection_lost, None)
  File "/Users/wei/opt/anaconda3/envs/autogen/lib/python3.11/asyncio/base_events.py", line 762, in call_soon
    self._check_closed()
  File "/Users/wei/opt/anaconda3/envs/autogen/lib/python3.11/asyncio/base_events.py", line 520, in _check_closed
    raise RuntimeError('Event loop is closed')
RuntimeError: Event loop is closed
INFO     [agent] 👍 Eval: Success - Google search results retrieved successfully for DeepSeek.
INFO     [agent] 🧠 Memory: Search result page loaded. 0/10 extracted. Begin extracting relevant content on DeepSeek's financial impact on stock prices.
INFO     [agent] 🎯 Next goal: Extract content from the listed search results about financial or economic factors impacting DeepSeek and its influence on stock prices.
INFO     [agent] 🛠️  Action 1/1: {"extract_content":{"goal":"financial or economic factors impacting DeepSeek and its influence on stock prices"}}
INFO     [controller] 📄  Extracted from page
...rest omitted

Eventually the tutorial run successfully, should I worry about this inner error?

[ww2283]

Question 2: Gemini model

When I use gemini-2.0-flash, the whole answer is significantly 'shallower' than gpt-4o, not 'deep' at all. Other gemini models will error out; e.g. the gemini-2.0-flash-lite-preview-02-05 or gemini-1.5-flash.

Below is the output

DeepResearchAgent (to user):


***** Suggested tool call (7764): delegate_research_task *****
Arguments: 
{"task": "impact of DeepSeek on stock prices and reasons for that"}
**************************************************************

--------------------------------------------------------------------------------

>>>>>>>> EXECUTING FUNCTION delegate_research_task...
Call ID: 7764
Input arguments: {'task': 'impact of DeepSeek on stock prices and reasons for that'}
CriticAgent (to SummarizerAgent):

Please answer the following question: impact of DeepSeek on stock prices and reasons for that

--------------------------------------------------------------------------------
INFO     [root] AFC is enabled with max remote calls: 10.
INFO     [root] AFC remote call 1 is done.
SummarizerAgent (to CriticAgent):


***** Suggested tool call (4654): split_question_and_answer_subquestions *****
Arguments: 
{"question": "impact of DeepSeek on stock prices and reasons for that"}
******************************************************************************

--------------------------------------------------------------------------------

>>>>>>>> EXECUTING FUNCTION split_question_and_answer_subquestions...
Call ID: 4654
Input arguments: {'question': 'impact of DeepSeek on stock prices and reasons for that'}
DecompositionCritic (to DecompositionAgent):

Analyse and gather subqestions for the following question: impact of DeepSeek on stock prices and reasons for that

--------------------------------------------------------------------------------
INFO     [root] AFC is enabled with max remote calls: 10.
INFO     [root] AFC remote call 1 is done.
DecompositionAgent (to DecompositionCritic):

1.  What is DeepSeek, and what are its primary functions/applications?
2.  What are the possible mechanisms through which DeepSeek could influence stock prices (e.g., news analysis, sentiment analysis, algorithmic trading)?
3.  What data sources would be needed to determine whether or not DeepSeek is correlated with stock prices?


--------------------------------------------------------------------------------
CriticAgent (to SummarizerAgent):

***** Response from calling tool (4654) *****
Error: 'type'
*********************************************

--------------------------------------------------------------------------------
INFO     [root] AFC is enabled with max remote calls: 10.
INFO     [root] AFC remote call 1 is done.
SummarizerAgent (to CriticAgent):

I am sorry, I cannot fulfill this request. The tool returned an error. Please try again.


--------------------------------------------------------------------------------
INFO     [root] AFC is enabled with max remote calls: 10.
INFO     [root] AFC remote call 1 is done.
user (to DeepResearchAgent):

***** Response from calling tool (7764) *****
Error: 'NoneType' object has no attribute 'parts'
*********************************************

--------------------------------------------------------------------------------

>>>>>>>> USING AUTO REPLY...
INFO     [root] AFC is enabled with max remote calls: 10.
INFO     [root] AFC remote call 1 is done.
DeepResearchAgent (to user):


***** Suggested tool call (1194): delegate_research_task *****
Arguments: 
{"task": "What was the impact of DeepSeek on stock prices and why?"}
**************************************************************

--------------------------------------------------------------------------------
INFO     [root] AFC is enabled with max remote calls: 10.
INFO     [root] AFC remote call 1 is done.
{'content': 'I am unable to summarize the conversation because I encountered an error when trying to research the impact of DeepSeek on stock prices.\n', 'refusal': None, 'role': 'assistant', 'audio': None, 'function_call': None, 'tool_calls': None}

[ww2283]

managed to print out error part around function call 4654 with critic agent:

Critic Agent Chat Result:
{
  "chat_id": null,
  "chat_history": [
    {
      "content": "Please answer the following question: impact of DeepSeek on stock prices and reasons for that",
      "role": "assistant",
      "name": "CriticAgent"
    },
    {
      "content": "",
      "tool_calls": [
        {
          "id": "4654",
          "function": {
            "arguments": "{\"question\": \"impact of DeepSeek on stock prices and reasons for that\"}",
            "name": "split_question_and_answer_subquestions"
          },
          "type": "function"
        }
      ],
      "role": "assistant"
    },
    {
      "content": "Error: 'type'",
      "tool_responses": [
        {
          "tool_call_id": "4654",
          "role": "tool",
          "content": "Error: 'type'"
        }
      ],
      "role": "tool",
      "name": "CriticAgent"
    },
    {
      "content": "I am sorry, I cannot fulfill this request. The tool returned an error. Please try again.\n",
      "role": "user",
      "name": "SummarizerAgent"
    },
    {
      "content": "",
      "tool_calls": [
        {
          "id": "2162",
          "function": {
            "arguments": "{\"answer\": \"The impact of DeepSeek on stock prices is difficult to assess directly. However, it could indirectly impact companies using DeepSeek (positive or negative based on successful integration), AI-related stocks (potentially positive due to increased confidence in the AI sector), and create speculative trading activity. A definitive assessment requires analyzing specific companies and their performance after adopting DeepSeek's technology, considering broader market trends.\", \"reasoning\": \"The answer provides a balanced assessment of potential indirect impacts of DeepSeek on stock prices, considering both positive and negative possibilities. It also acknowledges the speculative nature of some market reactions and highlights the need for specific data to make a definitive conclusion. The answer is coherent, relevant, and provides a reasonably complete overview given the limited information available about DeepSeek's market impact.\"}",
            "name": "confirm_summary"
          },
          "type": "function"
        }
      ],
      "role": "assistant"
    },
    {
      "content": "Answer confirmed:The impact of DeepSeek on stock prices is difficult to assess directly. However, it could indirectly impact companies using DeepSeek (positive or negative based on successful integration), AI-related stocks (potentially positive due to increased confidence in the AI sector), and create speculative trading activity. A definitive assessment requires analyzing specific companies and their performance after adopting DeepSeek's technology, considering broader market trends.\nReasoning: The answer provides a balanced assessment of potential indirect impacts of DeepSeek on stock prices, considering both positive and negative possibilities. It also acknowledges the speculative nature of some market reactions and highlights the need for specific data to make a definitive conclusion. The answer is coherent, relevant, and provides a reasonably complete overview given the limited information available about DeepSeek's market impact.",
      "tool_responses": [
        {
          "tool_call_id": "2162",
          "role": "tool",
          "content": "Answer confirmed:The impact of DeepSeek on stock prices is difficult to assess directly. However, it could indirectly impact companies using DeepSeek (positive or negative based on successful integration), AI-related stocks (potentially positive due to increased confidence in the AI sector), and create speculative trading activity. A definitive assessment requires analyzing specific companies and their performance after adopting DeepSeek's technology, considering broader market trends.\nReasoning: The answer provides a balanced assessment of potential indirect impacts of DeepSeek on stock prices, considering both positive and negative possibilities. It also acknowledges the speculative nature of some market reactions and highlights the need for specific data to make a definitive conclusion. The answer is coherent, relevant, and provides a reasonably complete overview given the limited information available about DeepSeek's market impact."
        }
      ],
      "role": "tool",
      "name": "SummarizerAgent"
    }
  ],
  "summary": "Answer confirmed:The impact of DeepSeek on stock prices is difficult to assess directly. However, it could indirectly impact companies using DeepSeek (positive or negative based on successful integration), AI-related stocks (potentially positive due to increased confidence in the AI sector), and create speculative trading activity. A definitive assessment requires analyzing specific companies and their performance after adopting DeepSeek's technology, considering broader market trends.\nReasoning: The answer provides a balanced assessment of potential indirect impacts of DeepSeek on stock prices, considering both positive and negative possibilities. It also acknowledges the speculative nature of some market reactions and highlights the need for specific data to make a definitive conclusion. The answer is coherent, relevant, and provides a reasonably complete overview given the limited information available about DeepSeek's market impact.",
  "cost": {
    "usage_including_cached_inference": {
      "total_cost": 0,
      "gemini-2.0-flash": {
        "cost": 0,
        "prompt_tokens": 426,
        "completion_tokens": 621,
        "total_tokens": 1047
      }
    },
    "usage_excluding_cached_inference": {
      "total_cost": 0
    }
  },
  "human_input": []
}

[ww2283]

added bug report at : #984

[qingyun-wu]

Thanks @ww2283 for the bug report! @sternakt, @rjambrecic and @alecsolder could you take a look? Thanks!