run-llama/llama_index · error · ValueError

code_execute_fn must be provided for CodeActAgent

Error message

code_execute_fn must be provided for CodeActAgent

What it means

QueryEngineTool._get_query_str derives the query string from either the first positional arg or the 'input' kwarg (matching the default fn_schema). If neither is present and _resolve_input_errors is False, it raises this ValueError — the tool was called without the input the query engine needs.

Source

Thrown at llama-index-core/llama_index/core/agent/workflow/codeact_agent.py:271

                    current_agent_name=self.name,
                    thinking_delta=last_chat_response.additional_kwargs.get(
                        "thinking_delta", None
                    ),
                )
            )

        return last_chat_response, full_response_text

    async def take_step(
        self,
        ctx: AgentContext,
        llm_input: List[ChatMessage],
        tools: Sequence[BaseTool],
        memory: BaseMemory,
    ) -> AgentOutput:
        """Take a single step with the code act agent."""
        if not self.code_execute_fn:
            raise ValueError("code_execute_fn must be provided for CodeActAgent")

        # Get current scratchpad
        scratchpad: List[ChatMessage] = await ctx.store.get(
            self.scratchpad_key, default=[]
        )
        current_llm_input = [*llm_input, *scratchpad]

        # Create a system message with tool descriptions
        tool_descriptions = self._get_tool_descriptions(tools)
        system_prompt = self.code_act_system_prompt.format(
            tool_descriptions=tool_descriptions
        )

        # Add or overwrite system message
        has_system = False
        for i, msg in enumerate(current_llm_input):
            if msg.role.value == "system":
                current_llm_input[i] = ChatMessage(role="system", content=system_prompt)

View on GitHub (pinned to afd0fef371)

Solutions

1. Pass the query as first positional or as input=: tool.call(input='What are the sales numbers?').
2. Construct the tool with resolve_input_errors=True (QueryEngineTool.from_defaults(..., resolve_input_errors=True)) so unexpected kwargs are stringified instead of raising.
3. Align the tool's fn_schema with how you call it: a schema with an input field.
4. Validate tool-call payloads before dispatch: ensure 'input' in kwargs or args non-empty.

Example fix

# before
tool = QueryEngineTool.from_defaults(query_engine=engine)
tool.call(query="sales numbers?")  # ValueError: Cannot call query engine without specifying `input`

# after
result = tool.call(input="sales numbers?")
# or make it tolerant:
# tool = QueryEngineTool.from_defaults(query_engine=engine, resolve_input_errors=True)

if not args and 'input' not in kwargs:
    if tool._resolve_input_errors:
        kwargs['input'] = str(kwargs)
    else:
        raise ValueError('missing input')
result = tool.call(*args, **kwargs)

def has_query_input(args: tuple, kwargs: dict) -> bool:
    return len(args) > 0 or 'input' in kwargs

try:
    out = tool.call(**llm_args)
except ValueError as e:
    if 'without specifying `input`' in str(e):
        out = tool.call(input=llm_args.get('query') or str(llm_args))
    else:
        raise

When it happens

Trigger: Calling query_engine_tool.call(**{'query': '...'}) (wrong kwarg name, e.g. 'query' instead of 'input'); an agent framework passing only non-positional metadata kwargs; fn_schema customized to a different field name while _get_query_str still expects input; _resolve_input_errors left False when the LLM produces malformed tool args.

Common situations: Hand-written agent loops that forward LLM JSON whose key is not 'input'; switching from OpenAIFunction agent (which uses input) to a custom runner; older tool metadata using 'query' key; LLM hallucinating argument names.

Related errors

• Max iterations of {max_iterations} reached! Either something
• No tool calls found, cannot aggregate results.
• Tool {tool.metadata.name} requires context. CodeActAgent onl
• Tool {tool.metadata.name} is not a FunctionTool. CodeActAgen
• LLM must be a FunctionCallingLLM

AI-assisted analysis of run-llama/llama_index@afd0fef371 (2026-08-15). Data as JSON: /api/errors/0e7a695fb2d4ee23. Report an issue: GitHub.