crewAIInc/crewAI · error · ValueError

The LlamaIndex tool does not have an fn_schema specified.

Error message

The LlamaIndex tool does not have an fn_schema specified.

What it means

After confirming the object is a llama-index BaseTool, LlamaIndexTool.from_tool requires tool.metadata.fn_schema (the pydantic schema describing the tool's arguments) because CrewAI builds its args_schema from it. Tools created without a schema — e.g. FunctionTool.from_defaults(fn=...) without explicit fn_schema on some paths, or custom tools with metadata.fn_schema left None — fail here.

Source

Thrown at lib/crewai-tools/src/crewai_tools/tools/llamaindex_tool/llamaindex_tool.py:37

        """Run tool."""
        tool = self.llama_index_tool

        if self.result_as_answer:
            return tool(*args, **kwargs).content

        return tool(*args, **kwargs)

    @classmethod
    def from_tool(cls, tool: Any, **kwargs: Any) -> LlamaIndexTool:
        from llama_index.core.tools import (  # type: ignore[import-not-found]
            BaseTool as LlamaBaseTool,
        )

        if not isinstance(tool, LlamaBaseTool):
            raise ValueError(f"Expected a LlamaBaseTool, got {type(tool)}")

        if tool.metadata.fn_schema is None:
            raise ValueError(
                "The LlamaIndex tool does not have an fn_schema specified."
            )
        args_schema = cast(type[BaseModel], tool.metadata.fn_schema)

        return cls(
            name=tool.metadata.name,
            description=tool.metadata.description,
            args_schema=args_schema,
            llama_index_tool=tool,
            **kwargs,
        )

    @classmethod
    def from_query_engine(
        cls,
        query_engine: Any,
        name: str | None = None,
        description: str | None = None,

View on GitHub (pinned to 754d7323be)

Solutions

1. Provide an explicit schema when building the llama-index tool: FunctionTool.from_defaults(fn=my_fn, fn_schema=MySchema) where MySchema is a pydantic BaseModel
2. Set tool.metadata.fn_schema manually before calling from_tool if you control the metadata
3. Upgrade llama-index so from_defaults reliably attaches DefaultToolFnSchema

Example fix

# before
from llama_index.core.tools import FunctionTool
t = FunctionTool.from_defaults(fn=lambda q: lookup(q))  # fn_schema may be None
crew_tool = LlamaIndexTool.from_tool(t)  # ValueError

# after
from pydantic import BaseModel, Field
class LookupSchema(BaseModel):
    query: str = Field(..., description="Query string")
t = FunctionTool.from_defaults(fn=lookup, fn_schema=LookupSchema)
crew_tool = LlamaIndexTool.from_tool(t)

def tool_has_fn_schema(tool) -> bool:
    meta = getattr(tool, "metadata", None)
    return getattr(meta, "fn_schema", None) is not None

def has_fn_schema(tool: Any) -> "TypeGuard[Any]":
    return getattr(getattr(tool, "metadata", None), "fn_schema", None) is not None

try:
    crew_tool = LlamaIndexTool.from_tool(tool)
except ValueError as e:
    if "fn_schema" in str(e):
        from pydantic import BaseModel, Field
        class Schema(BaseModel):
            query: str = Field(...)
        tool.metadata.fn_schema = Schema
        crew_tool = LlamaIndexTool.from_tool(tool)
    else:
        raise

When it happens

Trigger: Calling from_tool on a BaseTool whose ToolMetadata was constructed without fn_schema; passing tools built by older llama-index factory methods that infer schemas lazily; passing a tool whose fn_schema is dynamically None (e.g. async tools without schema).

Common situations: Wrapping hand-written llama-index tools; llama-index version drift where from_defaults no longer auto-generates fn_schema for the object you hold; wrapping tools that only define metadata at call time.

Related errors

• Unable to read --definition path {definition_path}: {exc}
• Invalid data_type: '{raw_data_type}'. Valid values are: 'fil
• Directory does not exist: {source_ref}
• Website URL must be provided either during initialization or
• Expected a LlamaBaseTool, got {type(tool)}

AI-assisted analysis of crewAIInc/crewAI@754d7323be (2026-08-15). Data as JSON: /api/errors/47132b3bb0a3a44b. Report an issue: GitHub.