Is `chat` session supposed to be able to accept `tools` within the conversation / within a message?

[Mr-Ruben]

Is chat session supposed to be able to accept tools within the conversation / within a message?

# --- Code ---
chat=Chat(
    model="openai/gpt-4.1-mini",
    #tools=[make_joke], 
    tools=None, # This is the default value
)
chat(
    msg="Tell me a joke about 'Admin User' working in 'Jupyter notebooks'.",
    tools=[make_joke],
)

Returns TypeError: litellm.main.completion() got multiple values for keyword argument 'tools'

Which seems to me that tools is passed down, but then conflicts with tools=None which comes from Chat.

As it was not expected, it wasn't handled and causes the problem.

I don't see the need to have to specify the tools at the beginning of the session. That may be adding details/tools/context when it is not required/needed.

# --- Output ---
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[43], line 6
      1 chat=Chat(
      2     model="openai/gpt-4.1-mini",
      3     #tools=[make_joke], 
      4     tools=None,
      5 )
----> 6 chat(
      7     msg="Tell me a joke about 'Admin User' working in 'Jupyter notebooks'.",
      8     #tools=[make_joke],
      9     tools=[make_joke],
     10 )

File ~/.local/lib/python3.11/site-packages/lisette/core.py:307, in Chat.__call__(self, msg, prefill, temp, think, search, stream, max_steps, final_prompt, return_all, **kwargs)
    305 if stream: return result_gen              # streaming
    306 elif return_all: return list(result_gen)  # toolloop behavior
--> 307 else: return last(result_gen)

File ~/.local/lib/python3.11/site-packages/fastcore/basics.py:749, in last(x, f, negate, **kwargs)
    747 if f: x = filter_ex(x, f=f, negate=negate, gen=True, **kwargs)
    748 res = None
--> 749 for res in x: pass
    750 return res

File ~/.local/lib/python3.11/site-packages/lisette/core.py:269, in Chat._call(self, msg, prefill, temp, think, search, stream, max_steps, step, final_prompt, tool_choice, **kwargs)
    267 if _has_search(self.model) and (s:=ifnone(search,self.search)): kwargs['web_search_options'] = {"search_context_size": effort[s]}
    268 else: _=kwargs.pop('web_search_options',None)
--> 269 res = completion(model=self.model, messages=self._prep_msg(msg, prefill), stream=stream, 
    270                  tools=self.tool_schemas, reasoning_effort = effort.get(think), tool_choice=tool_choice,
    271                  # temperature is not supported when reasoning
    272                  temperature=None if think else ifnone(temp,self.temp),
    273                  **kwargs)
    274 if stream:
    275     if prefill: yield _mk_prefill(prefill)

TypeError: litellm.main.completion() got multiple values for keyword argument 'tools'

Regards, R.

[Mr-Ruben]

I found a workaround

Creating this helper

def set_chat_tools(chat: Chat, tools: list) -> None:
    """
    v1: Set tools on an existing Chat, using the same logic as Chat.__init__.
    """
    from toolslm.funccall import mk_ns
    
    tools = listify(tools)  # Same as __init__
    chat.tools = tools
    chat.tool_schemas = [lisette.lite_mk_func(t) for t in tools] if tools else None
    chat.ns = mk_ns(tools) if tools else {}

I can insert the tools when needed

# Session creation
chat = Chat(model="openai/gpt-4.1-mini",)

# As we will make a call with tools, we need to set them
set_chat_tools(chat=chat, tools=[make_joke])
# Make the call
chat(
    msg="Tell me a joke about 'Admin User (Lukas)' working in 'Jupyter notebooks'.",
)

# --- Code ---
chat.hist


# --- Output ---
[{'role': 'user',
  'content': "Tell me a joke about 'Admin User (Lukas)' working in 'Jupyter notebooks'."},
 Message(content=None, role='assistant', tool_calls=[{'function': {'arguments': '{"name":"Admin User (Lukas)","topic":"Jupyter notebooks"}', 'name': 'make_joke'}, 'id': 'call_ECXQG0zuHHk2OPdMNoMbB9Rl', 'type': 'function'}], function_call=None, provider_specific_fields={'refusal': None}, annotations=[]),
 {'tool_call_id': 'call_ECXQG0zuHHk2OPdMNoMbB9Rl',
  'role': 'tool',
  'name': 'make_joke',
  'content': 'Why did Admin User (Lukas) bring a ladder to the Jupyter notebooks?\nBecause Admin User (Lukas) wanted to reach the highest cell outputs (in the 4.5 version of JupyterLab)!'},
 {'role': 'assistant',
  'content': 'You have no more tool uses. Please summarize your findings. If you did not complete your goal please tell the user what further work needs to be done so they can choose how best to proceed.'},
 Message(content="Here's a joke for you: \n\nWhy did Admin User (Lukas) bring a ladder to the Jupyter notebooks? Because Admin User (Lukas) wanted to reach the highest cell outputs (in the 4.5 version of JupyterLab)! \n\nIf you'd like another joke or something else, just let me know!", role='assistant', tool_calls=None, function_call=None, provider_specific_fields={'refusal': None}, annotations=[])]

Ideally, a change in chat.tools would trigger the actions that set_chat_tools does. So one could add/remove/reset the tools available at any time.

Maybe tomorrow I'll change my mind... :o)

[jph00]

Does this issue still repro @Mr-Ruben ?

[Mr-Ruben]

Does this issue still repro
Yes.

# install lisette
!pip install lisette


# show present version
!pip show lisette


# Name: lisette
# Version: 0.1.10
# Summary: litellm helper
# Home-page: 
# Author: 
# Author-email: AnswerDotAI <support@answer.ai>
# License: Apache-2.0
# Location: /usr/local/lib/python3.12/dist-packages
# Requires: fastcore, litellm, numpydoc, toolslm
# Required-by: 



# ....

Chat = lisette.Chat

# Here is my tool
def make_joke(name: str, topic: str) -> str:
    """Generate a joke about a person and a topic."""
    
    return f"""Why did {name} bring a ladder to the {topic}?
Because {name} wanted to reach the highest cell outputs.""" 






chat = Chat(model="openai/gpt-4.1-mini")

# Make the call
chat(msg="Tell me a joke about 'Admin User (Lukas)' working in 'Jupyter notebooks'.",
     tools=[make_joke],
     )

Output

 ---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
/tmp/ipykernel_6577/759078584.py in <cell line: 0>()
      2 
      3 # Make the call
----> 4 chat(
      5     msg="Tell me a joke about 'Admin User (Lukas)' working in 'Jupyter notebooks'.",
      6      tools=[make_joke],

2 frames/usr/local/lib/python3.12/dist-packages/lisette/core.py in __call__(self, msg, prefill, temp, think, search, stream, max_steps, final_prompt, return_all, **kwargs)
    591     if stream: return result_gen              # streaming
    592     elif return_all: return list(result_gen)  # toolloop behavior
--> 593     else: return last(result_gen)             # normal chat behavior
    594 
    595 # %% ../nbs/00_core.ipynb #2e9247ba

/usr/local/lib/python3.12/dist-packages/fastcore/basics.py in last(x, f, negate, **kwargs)
    767     if f: x = filter_ex(x, f=f, negate=negate, gen=True, **kwargs)
    768     res = None
--> 769     for res in x: pass
    770     return res
    771 

/usr/local/lib/python3.12/dist-packages/lisette/core.py in _call(self, msg, prefill, temp, think, search, stream, max_steps, step, final_prompt, tool_choice, max_tokens, **kwargs)
    535     if step>max_steps+1: return
    536     prefill, max_tokens = self._prep_call(prefill, search, max_tokens, kwargs, stream=stream)
--> 537     res = completion(
    538         model=self.model, messages=self._prep_msg(msg, prefill), stream=stream, max_tokens=int(max_tokens),
    539         tools=self.tool_schemas, reasoning_effort = effort.get(think), tool_choice=tool_choice, num_retries=2,

TypeError: litellm.main.completion() got multiple values for keyword argument 'tools'