17378114148

Committed 01 Sep 2025 12:48PM UTC coverage: 92.077% (-0.004%) from 92.081%

Build # 17378114148

Build Type

push

github

Committed by

web-flow

Commit Message

refactor: Some refactoring and finish TODO for component tool (#9751)

* Some refactoring and finish TODO for component tool

* Add docstrings

Run Details

12935 of 14048 relevant lines covered (92.08%)

0.92 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

93.98

haystack/tools/component_tool.py

# SPDX-FileCopyrightText: 2022-present deepset GmbH <info@deepset.ai>
#
# SPDX-License-Identifier: Apache-2.0

from typing import Any, Callable, Optional, Union, get_args, get_origin

from pydantic import Field, TypeAdapter, create_model

from haystack import logging
from haystack.core.component import Component
from haystack.core.serialization import (
    component_from_dict,
    component_to_dict,
    generate_qualified_class_name,
    import_class_by_name,
)
from haystack.tools import Tool
from haystack.tools.errors import SchemaGenerationError
from haystack.tools.from_function import _remove_title_from_schema
from haystack.tools.parameters_schema_utils import _get_component_param_descriptions, _resolve_type
from haystack.tools.tool import _deserialize_outputs_to_state, _serialize_outputs_to_state
from haystack.utils.callable_serialization import deserialize_callable, serialize_callable

logger = logging.getLogger(__name__)


class ComponentTool(Tool):
    """
    A Tool that wraps Haystack components, allowing them to be used as tools by LLMs.

    ComponentTool automatically generates LLM-compatible tool schemas from component input sockets,
    which are derived from the component's `run` method signature and type hints.


    Key features:
    - Automatic LLM tool calling schema generation from component input sockets
    - Type conversion and validation for component inputs
    - Support for types:
        - Dataclasses
        - Lists of dataclasses
        - Basic types (str, int, float, bool, dict)
        - Lists of basic types
    - Automatic name generation from component class name
    - Description extraction from component docstrings

    To use ComponentTool, you first need a Haystack component - either an existing one or a new one you create.
    You can create a ComponentTool from the component by passing the component to the ComponentTool constructor.
    Below is an example of creating a ComponentTool from an existing SerperDevWebSearch component.

    ## Usage Example:

    ```python
    from haystack import component, Pipeline
    from haystack.tools import ComponentTool
    from haystack.components.websearch import SerperDevWebSearch
    from haystack.utils import Secret
    from haystack.components.tools.tool_invoker import ToolInvoker
    from haystack.components.generators.chat import OpenAIChatGenerator
    from haystack.dataclasses import ChatMessage

    # Create a SerperDev search component
    search = SerperDevWebSearch(api_key=Secret.from_env_var("SERPERDEV_API_KEY"), top_k=3)

    # Create a tool from the component
    tool = ComponentTool(
        component=search,
        name="web_search",  # Optional: defaults to "serper_dev_web_search"
        description="Search the web for current information on any topic"  # Optional: defaults to component docstring
    )

    # Create pipeline with OpenAIChatGenerator and ToolInvoker
    pipeline = Pipeline()
    pipeline.add_component("llm", OpenAIChatGenerator(model="gpt-4o-mini", tools=[tool]))
    pipeline.add_component("tool_invoker", ToolInvoker(tools=[tool]))

    # Connect components
    pipeline.connect("llm.replies", "tool_invoker.messages")

    message = ChatMessage.from_user("Use the web search tool to find information about Nikola Tesla")

    # Run pipeline
    result = pipeline.run({"llm": {"messages": [message]}})

    print(result)
    ```

    """

    def __init__(
        self,
        component: Component,
        name: Optional[str] = None,
        description: Optional[str] = None,
        parameters: Optional[dict[str, Any]] = None,
        *,
        outputs_to_string: Optional[dict[str, Union[str, Callable[[Any], str]]]] = None,
        inputs_from_state: Optional[dict[str, str]] = None,
        outputs_to_state: Optional[dict[str, dict[str, Union[str, Callable]]]] = None,
    ) -> None:
        """
        Create a Tool instance from a Haystack component.

        :param component: The Haystack component to wrap as a tool.
        :param name: Optional name for the tool (defaults to snake_case of component class name).
        :param description: Optional description (defaults to component's docstring).
        :param parameters:
            A JSON schema defining the parameters expected by the Tool.
            Will fall back to the parameters defined in the component's run method signature if not provided.
        :param outputs_to_string:
            Optional dictionary defining how a tool outputs should be converted into a string.
            If the source is provided only the specified output key is sent to the handler.
            If the source is omitted the whole tool result is sent to the handler.
            Example:
            ```python
            {
                "source": "docs", "handler": format_documents
            }
            ```
        :param inputs_from_state:
            Optional dictionary mapping state keys to tool parameter names.
            Example: `{"repository": "repo"}` maps state's "repository" to tool's "repo" parameter.
        :param outputs_to_state:
            Optional dictionary defining how tool outputs map to keys within state as well as optional handlers.
            If the source is provided only the specified output key is sent to the handler.
            Example:
            ```python
            {
                "documents": {"source": "docs", "handler": custom_handler}
            }
            ```
            If the source is omitted the whole tool result is sent to the handler.
            Example:
            ```python
            {
                "documents": {"handler": custom_handler}
            }
            ```
        :raises ValueError: If the component is invalid or schema generation fails.
        """
        if not isinstance(component, Component):
            message = (
                f"Object {component!r} is not a Haystack component. "
                "Use ComponentTool only with Haystack component instances."
            )
            raise ValueError(message)

        if getattr(component, "__haystack_added_to_pipeline__", None):
            msg = (
                "Component has been added to a pipeline and can't be used to create a ComponentTool. "
                "Create ComponentTool from a non-pipeline component instead."
            )
            raise ValueError(msg)

        self._unresolved_parameters = parameters
        # Create the tools schema from the component run method parameters
        tool_schema = parameters or self._create_tool_parameters_schema(component, inputs_from_state or {})

        def component_invoker(**kwargs):
            """
            Invokes the component using keyword arguments provided by the LLM function calling/tool-generated response.

            :param kwargs: The keyword arguments to invoke the component with.
            :returns: The result of the component invocation.
            """
            converted_kwargs = {}
            input_sockets = component.__haystack_input__._sockets_dict  # type: ignore[attr-defined]
            for param_name, param_value in kwargs.items():
                param_type = input_sockets[param_name].type

                # Check if the type (or list element type) has from_dict
                target_type = get_args(param_type)[0] if get_origin(param_type) is list else param_type
                if hasattr(target_type, "from_dict"):
                    if isinstance(param_value, list):
                        resolved_param_value = [
                            target_type.from_dict(item) if isinstance(item, dict) else item for item in param_value
                        ]
                    elif isinstance(param_value, dict):
                        resolved_param_value = target_type.from_dict(param_value)
                    else:
                        resolved_param_value = param_value
                else:
                    # Let TypeAdapter handle both single values and lists
                    type_adapter = TypeAdapter(param_type)
                    resolved_param_value = type_adapter.validate_python(param_value)

                converted_kwargs[param_name] = resolved_param_value
            logger.debug(f"Invoking component {type(component)} with kwargs: {converted_kwargs}")
            return component.run(**converted_kwargs)

        # Generate a name for the tool if not provided
        if not name:
            class_name = component.__class__.__name__
            # Convert camelCase/PascalCase to snake_case
            name = "".join(
                [
                    "_" + c.lower() if c.isupper() and i > 0 and not class_name[i - 1].isupper() else c.lower()
                    for i, c in enumerate(class_name)
                ]
            ).lstrip("_")

        description = description or component.__doc__ or name

        # Create the Tool instance with the component invoker as the function to be called and the schema
        super().__init__(
            name=name,
            description=description,
            parameters=tool_schema,
            function=component_invoker,
            inputs_from_state=inputs_from_state,
            outputs_to_state=outputs_to_state,
            outputs_to_string=outputs_to_string,
        )
        self._component = component

    def to_dict(self) -> dict[str, Any]:
        """
        Serializes the ComponentTool to a dictionary.
        """
        serialized: dict[str, Any] = {
            "component": component_to_dict(obj=self._component, name=self.name),
            "name": self.name,
            "description": self.description,
            "parameters": self._unresolved_parameters,
            "inputs_from_state": self.inputs_from_state,
            # This is soft-copied as to not modify the attributes in place
            "outputs_to_state": self.outputs_to_state.copy() if self.outputs_to_state else None,
        }

        if self.outputs_to_state is not None:
            serialized["outputs_to_state"] = _serialize_outputs_to_state(self.outputs_to_state)

        if self.outputs_to_string is not None and self.outputs_to_string.get("handler") is not None:
            # This is soft-copied as to not modify the attributes in place
            serialized["outputs_to_string"] = self.outputs_to_string.copy()
            serialized["outputs_to_string"]["handler"] = serialize_callable(self.outputs_to_string["handler"])
        else:
            serialized["outputs_to_string"] = None

        return {"type": generate_qualified_class_name(type(self)), "data": serialized}

    @classmethod
    def from_dict(cls, data: dict[str, Any]) -> "Tool":
        """
        Deserializes the ComponentTool from a dictionary.
        """
        inner_data = data["data"]
        component_class = import_class_by_name(inner_data["component"]["type"])
        component = component_from_dict(cls=component_class, data=inner_data["component"], name=inner_data["name"])

        if "outputs_to_state" in inner_data and inner_data["outputs_to_state"]:
            inner_data["outputs_to_state"] = _deserialize_outputs_to_state(inner_data["outputs_to_state"])

        if (
            inner_data.get("outputs_to_string") is not None
            and inner_data["outputs_to_string"].get("handler") is not None
        ):
            inner_data["outputs_to_string"]["handler"] = deserialize_callable(
                inner_data["outputs_to_string"]["handler"]
            )

        return cls(
            component=component,
            name=inner_data["name"],
            description=inner_data["description"],
            parameters=inner_data.get("parameters", None),
            outputs_to_string=inner_data.get("outputs_to_string", None),
            inputs_from_state=inner_data.get("inputs_from_state", None),
            outputs_to_state=inner_data.get("outputs_to_state", None),
        )

    def _create_tool_parameters_schema(self, component: Component, inputs_from_state: dict[str, Any]) -> dict[str, Any]:
        """
        Creates an OpenAI tools schema from a component's run method parameters.

        :param component: The component to create the schema from.
        :raises SchemaGenerationError: If schema generation fails
        :returns: OpenAI tools schema for the component's run method parameters.
        """
        component_run_description, param_descriptions = _get_component_param_descriptions(component)

        # collect fields (types and defaults) and descriptions from function parameters
        fields: dict[str, Any] = {}

        for input_name, socket in component.__haystack_input__._sockets_dict.items():  # type: ignore[attr-defined]
            if inputs_from_state is not None and input_name in inputs_from_state:
                continue
            input_type = socket.type
            description = param_descriptions.get(input_name, f"Input '{input_name}' for the component.")

            # if the parameter has not a default value, Pydantic requires an Ellipsis (...)
            # to explicitly indicate that the parameter is required
            default = ... if socket.is_mandatory else socket.default_value
            resolved_type = _resolve_type(input_type)
            fields[input_name] = (resolved_type, Field(default=default, description=description))

        parameters_schema: dict[str, Any] = {}
        try:
            model = create_model(component.run.__name__, __doc__=component_run_description, **fields)
            parameters_schema = model.model_json_schema()
        except Exception as e:
            raise SchemaGenerationError(
                f"Failed to create JSON schema for the run method of Component '{component.__class__.__name__}'"
            ) from e

        # we don't want to include title keywords in the schema, as they contain redundant information
        # there is no programmatic way to prevent Pydantic from adding them, so we remove them later
        # see https://github.com/pydantic/pydantic/discussions/8504
        _remove_title_from_schema(parameters_schema)

        return parameters_schema

1	# SPDX-FileCopyrightText: 2022-present deepset GmbH <info@deepset.ai>
2	#
3	# SPDX-License-Identifier: Apache-2.0
4
5	from typing import Any, Callable, Optional, Union, get_args, get_origin	1✔
6
7	from pydantic import Field, TypeAdapter, create_model	1✔
8
9	from haystack import logging	1✔
10	from haystack.core.component import Component	1✔
11	from haystack.core.serialization import (	1✔
12	component_from_dict,
13	component_to_dict,
14	generate_qualified_class_name,
15	import_class_by_name,
16	)
17	from haystack.tools import Tool	1✔
18	from haystack.tools.errors import SchemaGenerationError	1✔
19	from haystack.tools.from_function import _remove_title_from_schema	1✔
20	from haystack.tools.parameters_schema_utils import _get_component_param_descriptions, _resolve_type	1✔
21	from haystack.tools.tool import _deserialize_outputs_to_state, _serialize_outputs_to_state	1✔
22	from haystack.utils.callable_serialization import deserialize_callable, serialize_callable	1✔
23
24	logger = logging.getLogger(__name__)	1✔
25
26
27	class ComponentTool(Tool):	1✔
28	"""
29	A Tool that wraps Haystack components, allowing them to be used as tools by LLMs.
30
31	ComponentTool automatically generates LLM-compatible tool schemas from component input sockets,
32	which are derived from the component's `run` method signature and type hints.
33
34
35	Key features:
36	- Automatic LLM tool calling schema generation from component input sockets
37	- Type conversion and validation for component inputs
38	- Support for types:
39	- Dataclasses
40	- Lists of dataclasses
41	- Basic types (str, int, float, bool, dict)
42	- Lists of basic types
43	- Automatic name generation from component class name
44	- Description extraction from component docstrings
45
46	To use ComponentTool, you first need a Haystack component - either an existing one or a new one you create.
47	You can create a ComponentTool from the component by passing the component to the ComponentTool constructor.
48	Below is an example of creating a ComponentTool from an existing SerperDevWebSearch component.
49
50	## Usage Example:
51
52	```python
53	from haystack import component, Pipeline
54	from haystack.tools import ComponentTool
55	from haystack.components.websearch import SerperDevWebSearch
56	from haystack.utils import Secret
57	from haystack.components.tools.tool_invoker import ToolInvoker
58	from haystack.components.generators.chat import OpenAIChatGenerator
59	from haystack.dataclasses import ChatMessage
60
61	# Create a SerperDev search component
62	search = SerperDevWebSearch(api_key=Secret.from_env_var("SERPERDEV_API_KEY"), top_k=3)
63
64	# Create a tool from the component
65	tool = ComponentTool(
66	component=search,
67	name="web_search", # Optional: defaults to "serper_dev_web_search"
68	description="Search the web for current information on any topic" # Optional: defaults to component docstring
69	)
70
71	# Create pipeline with OpenAIChatGenerator and ToolInvoker
72	pipeline = Pipeline()
73	pipeline.add_component("llm", OpenAIChatGenerator(model="gpt-4o-mini", tools=[tool]))
74	pipeline.add_component("tool_invoker", ToolInvoker(tools=[tool]))
75
76	# Connect components
77	pipeline.connect("llm.replies", "tool_invoker.messages")
78
79	message = ChatMessage.from_user("Use the web search tool to find information about Nikola Tesla")
80
81	# Run pipeline
82	result = pipeline.run({"llm": {"messages": [message]}})
83
84	print(result)
85	```
86
87	"""
88
89	def __init__(	1✔
90	self,
91	component: Component,
92	name: Optional[str] = None,
93	description: Optional[str] = None,
94	parameters: Optional[dict[str, Any]] = None,
95	*,
96	outputs_to_string: Optional[dict[str, Union[str, Callable[[Any], str]]]] = None,
97	inputs_from_state: Optional[dict[str, str]] = None,
98	outputs_to_state: Optional[dict[str, dict[str, Union[str, Callable]]]] = None,
99	) -> None:
100	"""
101	Create a Tool instance from a Haystack component.
102
103	:param component: The Haystack component to wrap as a tool.
104	:param name: Optional name for the tool (defaults to snake_case of component class name).
105	:param description: Optional description (defaults to component's docstring).
106	:param parameters:
107	A JSON schema defining the parameters expected by the Tool.
108	Will fall back to the parameters defined in the component's run method signature if not provided.
109	:param outputs_to_string:
110	Optional dictionary defining how a tool outputs should be converted into a string.
111	If the source is provided only the specified output key is sent to the handler.
112	If the source is omitted the whole tool result is sent to the handler.
113	Example:
114	```python
115	{
116	"source": "docs", "handler": format_documents
117	}
118	```
119	:param inputs_from_state:
120	Optional dictionary mapping state keys to tool parameter names.
121	Example: `{"repository": "repo"}` maps state's "repository" to tool's "repo" parameter.
122	:param outputs_to_state:
123	Optional dictionary defining how tool outputs map to keys within state as well as optional handlers.
124	If the source is provided only the specified output key is sent to the handler.
125	Example:
126	```python
127	{
128	"documents": {"source": "docs", "handler": custom_handler}
129	}
130	```
131	If the source is omitted the whole tool result is sent to the handler.
132	Example:
133	```python
134	{
135	"documents": {"handler": custom_handler}
136	}
137	```
138	:raises ValueError: If the component is invalid or schema generation fails.
139	"""
140	if not isinstance(component, Component):	1✔
141	message = (	1✔
142	f"Object {component!r} is not a Haystack component. "
143	"Use ComponentTool only with Haystack component instances."
144	)
145	raise ValueError(message)	1✔
146
147	if getattr(component, "__haystack_added_to_pipeline__", None):	1✔
148	msg = (	1✔
149	"Component has been added to a pipeline and can't be used to create a ComponentTool. "
150	"Create ComponentTool from a non-pipeline component instead."
151	)
152	raise ValueError(msg)	1✔
153
154	self._unresolved_parameters = parameters	1✔
155	# Create the tools schema from the component run method parameters
156	tool_schema = parameters or self._create_tool_parameters_schema(component, inputs_from_state or {})	1✔
157
158	def component_invoker(**kwargs):	1✔
159	"""
160	Invokes the component using keyword arguments provided by the LLM function calling/tool-generated response.
161
162	:param kwargs: The keyword arguments to invoke the component with.
163	:returns: The result of the component invocation.
164	"""
165	converted_kwargs = {}	1✔
166	input_sockets = component.__haystack_input__._sockets_dict # type: ignore[attr-defined]	1✔
167	for param_name, param_value in kwargs.items():	1✔
168	param_type = input_sockets[param_name].type	1✔
169
170	# Check if the type (or list element type) has from_dict
171	target_type = get_args(param_type)[0] if get_origin(param_type) is list else param_type	1✔
172	if hasattr(target_type, "from_dict"):	1✔
173	if isinstance(param_value, list):	1✔
174	resolved_param_value = [	1✔
175	target_type.from_dict(item) if isinstance(item, dict) else item for item in param_value
176	]
177	elif isinstance(param_value, dict):	×
178	resolved_param_value = target_type.from_dict(param_value)	×
179	else:
180	resolved_param_value = param_value	×
181	else:
182	# Let TypeAdapter handle both single values and lists
183	type_adapter = TypeAdapter(param_type)	1✔
184	resolved_param_value = type_adapter.validate_python(param_value)	1✔
185
186	converted_kwargs[param_name] = resolved_param_value	1✔
187	logger.debug(f"Invoking component {type(component)} with kwargs: {converted_kwargs}")	1✔
188	return component.run(**converted_kwargs)	1✔
189
190	# Generate a name for the tool if not provided
191	if not name:	1✔
192	class_name = component.__class__.__name__	1✔
193	# Convert camelCase/PascalCase to snake_case
194	name = "".join(	1✔
195	[
196	"_" + c.lower() if c.isupper() and i > 0 and not class_name[i - 1].isupper() else c.lower()
197	for i, c in enumerate(class_name)
198	]
199	).lstrip("_")
200
201	description = description or component.__doc__ or name	1✔
202
203	# Create the Tool instance with the component invoker as the function to be called and the schema
204	super().__init__(	1✔
205	name=name,
206	description=description,
207	parameters=tool_schema,
208	function=component_invoker,
209	inputs_from_state=inputs_from_state,
210	outputs_to_state=outputs_to_state,
211	outputs_to_string=outputs_to_string,
212	)
213	self._component = component	1✔
214
215	def to_dict(self) -> dict[str, Any]:	1✔
216	"""
217	Serializes the ComponentTool to a dictionary.
218	"""
219	serialized: dict[str, Any] = {	1✔
220	"component": component_to_dict(obj=self._component, name=self.name),
221	"name": self.name,
222	"description": self.description,
223	"parameters": self._unresolved_parameters,
224	"inputs_from_state": self.inputs_from_state,
225	# This is soft-copied as to not modify the attributes in place
226	"outputs_to_state": self.outputs_to_state.copy() if self.outputs_to_state else None,
227	}
228
229	if self.outputs_to_state is not None:	1✔
230	serialized["outputs_to_state"] = _serialize_outputs_to_state(self.outputs_to_state)	1✔
231
232	if self.outputs_to_string is not None and self.outputs_to_string.get("handler") is not None:	1✔
233	# This is soft-copied as to not modify the attributes in place
234	serialized["outputs_to_string"] = self.outputs_to_string.copy()	1✔
235	serialized["outputs_to_string"]["handler"] = serialize_callable(self.outputs_to_string["handler"])	1✔
236	else:
237	serialized["outputs_to_string"] = None	1✔
238
239	return {"type": generate_qualified_class_name(type(self)), "data": serialized}	1✔
240
241	@classmethod	1✔
242	def from_dict(cls, data: dict[str, Any]) -> "Tool":	1✔
243	"""
244	Deserializes the ComponentTool from a dictionary.
245	"""
246	inner_data = data["data"]	1✔
247	component_class = import_class_by_name(inner_data["component"]["type"])	1✔
248	component = component_from_dict(cls=component_class, data=inner_data["component"], name=inner_data["name"])	1✔
249
250	if "outputs_to_state" in inner_data and inner_data["outputs_to_state"]:	1✔
251	inner_data["outputs_to_state"] = _deserialize_outputs_to_state(inner_data["outputs_to_state"])	1✔
252
253	if (	1✔
254	inner_data.get("outputs_to_string") is not None
255	and inner_data["outputs_to_string"].get("handler") is not None
256	):
257	inner_data["outputs_to_string"]["handler"] = deserialize_callable(	1✔
258	inner_data["outputs_to_string"]["handler"]
259	)
260
261	return cls(	1✔
262	component=component,
263	name=inner_data["name"],
264	description=inner_data["description"],
265	parameters=inner_data.get("parameters", None),
266	outputs_to_string=inner_data.get("outputs_to_string", None),
267	inputs_from_state=inner_data.get("inputs_from_state", None),
268	outputs_to_state=inner_data.get("outputs_to_state", None),
269	)
270
271	def _create_tool_parameters_schema(self, component: Component, inputs_from_state: dict[str, Any]) -> dict[str, Any]:	1✔
272	"""
273	Creates an OpenAI tools schema from a component's run method parameters.
274
275	:param component: The component to create the schema from.
276	:raises SchemaGenerationError: If schema generation fails
277	:returns: OpenAI tools schema for the component's run method parameters.
278	"""
279	component_run_description, param_descriptions = _get_component_param_descriptions(component)	1✔
280
281	# collect fields (types and defaults) and descriptions from function parameters
282	fields: dict[str, Any] = {}	1✔
283
284	for input_name, socket in component.__haystack_input__._sockets_dict.items(): # type: ignore[attr-defined]	1✔
285	if inputs_from_state is not None and input_name in inputs_from_state:	1✔
286	continue	1✔
287	input_type = socket.type	1✔
288	description = param_descriptions.get(input_name, f"Input '{input_name}' for the component.")	1✔
289
290	# if the parameter has not a default value, Pydantic requires an Ellipsis (...)
291	# to explicitly indicate that the parameter is required
292	default = ... if socket.is_mandatory else socket.default_value	1✔
293	resolved_type = _resolve_type(input_type)	1✔
294	fields[input_name] = (resolved_type, Field(default=default, description=description))	1✔
295
296	parameters_schema: dict[str, Any] = {}	1✔
297	try:	1✔
298	model = create_model(component.run.__name__, __doc__=component_run_description, **fields)	1✔
299	parameters_schema = model.model_json_schema()	1✔
300	except Exception as e:	×
301	raise SchemaGenerationError(	×
302	f"Failed to create JSON schema for the run method of Component '{component.__class__.__name__}'"
303	) from e
304
305	# we don't want to include title keywords in the schema, as they contain redundant information
306	# there is no programmatic way to prevent Pydantic from adding them, so we remove them later
307	# see https://github.com/pydantic/pydantic/discussions/8504
308	_remove_title_from_schema(parameters_schema)	1✔
309
310	return parameters_schema	1✔

deepset-ai / haystack / 17378114148

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous