pydantic-ai-slim 1.0.8__tar.gz → 1.0.10__tar.gz

{pydantic_ai_slim-1.0.8 → pydantic_ai_slim-1.0.10}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pydantic-ai-slim
-Version: 1.0.8
+Version: 1.0.10
 Summary: Agent Framework / shim to use Pydantic with LLMs, slim package
 Project-URL: Homepage, https://github.com/pydantic/pydantic-ai/tree/main/pydantic_ai_slim
 Project-URL: Source, https://github.com/pydantic/pydantic-ai/tree/main/pydantic_ai_slim
@@ -33,7 +33,7 @@ Requires-Dist: genai-prices>=0.0.23
 Requires-Dist: griffe>=1.3.2
 Requires-Dist: httpx>=0.27
 Requires-Dist: opentelemetry-api>=1.28.0
-Requires-Dist: pydantic-graph==1.0.8
+Requires-Dist: pydantic-graph==1.0.10
 Requires-Dist: pydantic>=2.10
 Requires-Dist: typing-inspection>=0.4.0
 Provides-Extra: a2a
@@ -53,11 +53,11 @@ Requires-Dist: rich>=13; extra == 'cli'
 Provides-Extra: cohere
 Requires-Dist: cohere>=5.18.0; (platform_system != 'Emscripten') and extra == 'cohere'
 Provides-Extra: dbos
-Requires-Dist: dbos>=1.13.0; extra == 'dbos'
+Requires-Dist: dbos>=1.14.0; extra == 'dbos'
 Provides-Extra: duckduckgo
 Requires-Dist: ddgs>=9.0.0; extra == 'duckduckgo'
 Provides-Extra: evals
-Requires-Dist: pydantic-evals==1.0.8; extra == 'evals'
+Requires-Dist: pydantic-evals==1.0.10; extra == 'evals'
 Provides-Extra: google
 Requires-Dist: google-genai>=1.31.0; extra == 'google'
 Provides-Extra: groq
@@ -71,7 +71,7 @@ Requires-Dist: mcp>=1.12.3; extra == 'mcp'
 Provides-Extra: mistral
 Requires-Dist: mistralai>=1.9.10; extra == 'mistral'
 Provides-Extra: openai
-Requires-Dist: openai>=1.99.9; extra == 'openai'
+Requires-Dist: openai>=1.107.2; extra == 'openai'
 Provides-Extra: retries
 Requires-Dist: tenacity>=8.2.3; extra == 'retries'
 Provides-Extra: tavily

{pydantic_ai_slim-1.0.8 → pydantic_ai_slim-1.0.10}/pydantic_ai/_agent_graph.py

@@ -545,23 +545,26 @@ class CallToolsNode(AgentNode[DepsT, NodeRunEndT]):
             # Ensure that the stream is only run once
 
             async def _run_stream() -> AsyncIterator[_messages.HandleResponseEvent]:  # noqa: C901
-                texts: list[str] = []
+                text = ''
                 tool_calls: list[_messages.ToolCallPart] = []
-                thinking_parts: list[_messages.ThinkingPart] = []
+                invisible_parts: bool = False
 
                 for part in self.model_response.parts:
                     if isinstance(part, _messages.TextPart):
-                        # ignore empty content for text parts, see #437
-                        if part.content:
-                            texts.append(part.content)
+                        text += part.content
                     elif isinstance(part, _messages.ToolCallPart):
                         tool_calls.append(part)
                     elif isinstance(part, _messages.BuiltinToolCallPart):
-                        yield _messages.BuiltinToolCallEvent(part)
+                        # Text parts before a built-in tool call are essentially thoughts,
+                        # not part of the final result output, so we reset the accumulated text
+                        text = ''
+                        invisible_parts = True
+                        yield _messages.BuiltinToolCallEvent(part)  # pyright: ignore[reportDeprecated]
                     elif isinstance(part, _messages.BuiltinToolReturnPart):
-                        yield _messages.BuiltinToolResultEvent(part)
+                        invisible_parts = True
+                        yield _messages.BuiltinToolResultEvent(part)  # pyright: ignore[reportDeprecated]
                     elif isinstance(part, _messages.ThinkingPart):
-                        thinking_parts.append(part)
+                        invisible_parts = True
                     else:
                         assert_never(part)
 
@@ -569,36 +572,51 @@ class CallToolsNode(AgentNode[DepsT, NodeRunEndT]):
                 # In the future, we'd consider making this configurable at the agent or run level.
                 # This accounts for cases like anthropic returns that might contain a text response
                 # and a tool call response, where the text response just indicates the tool call will happen.
-                if tool_calls:
-                    async for event in self._handle_tool_calls(ctx, tool_calls):
-                        yield event
-                elif texts:
-                    # No events are emitted during the handling of text responses, so we don't need to yield anything
-                    self._next_node = await self._handle_text_response(ctx, texts)
-                elif thinking_parts:
-                    # handle thinking-only responses (responses that contain only ThinkingPart instances)
-                    # this can happen with models that support thinking mode when they don't provide
-                    # actionable output alongside their thinking content.
-                    self._next_node = ModelRequestNode[DepsT, NodeRunEndT](
-                        _messages.ModelRequest(
-                            parts=[_messages.RetryPromptPart('Responses without text or tool calls are not permitted.')]
+                try:
+                    if tool_calls:
+                        async for event in self._handle_tool_calls(ctx, tool_calls):
+                            yield event
+                    elif text:
+                        # No events are emitted during the handling of text responses, so we don't need to yield anything
+                        self._next_node = await self._handle_text_response(ctx, text)
+                    elif invisible_parts:
+                        # handle responses with only thinking or built-in tool parts.
+                        # this can happen with models that support thinking mode when they don't provide
+                        # actionable output alongside their thinking content. so we tell the model to try again.
+                        m = _messages.RetryPromptPart(
+                            content='Responses without text or tool calls are not permitted.',
                         )
-                    )
-                else:
-                    # we got an empty response with no tool calls, text, or thinking
-                    # this sometimes happens with anthropic (and perhaps other models)
-                    # when the model has already returned text along side tool calls
-                    # in this scenario, if text responses are allowed, we return text from the most recent model
-                    # response, if any
-                    if isinstance(ctx.deps.output_schema, _output.TextOutputSchema):
-                        for message in reversed(ctx.state.message_history):
-                            if isinstance(message, _messages.ModelResponse):
-                                last_texts = [p.content for p in message.parts if isinstance(p, _messages.TextPart)]
-                                if last_texts:
-                                    self._next_node = await self._handle_text_response(ctx, last_texts)
-                                    return
-
-                    raise exceptions.UnexpectedModelBehavior('Received empty model response')
+                        raise ToolRetryError(m)
+                    else:
+                        # we got an empty response with no tool calls, text, thinking, or built-in tool calls.
+                        # this sometimes happens with anthropic (and perhaps other models)
+                        # when the model has already returned text along side tool calls
+                        # in this scenario, if text responses are allowed, we return text from the most recent model
+                        # response, if any
+                        if isinstance(ctx.deps.output_schema, _output.TextOutputSchema):
+                            for message in reversed(ctx.state.message_history):
+                                if isinstance(message, _messages.ModelResponse):
+                                    text = ''
+                                    for part in message.parts:
+                                        if isinstance(part, _messages.TextPart):
+                                            text += part.content
+                                        elif isinstance(part, _messages.BuiltinToolCallPart):
+                                            # Text parts before a built-in tool call are essentially thoughts,
+                                            # not part of the final result output, so we reset the accumulated text
+                                            text = ''  # pragma: no cover
+                                    if text:
+                                        self._next_node = await self._handle_text_response(ctx, text)
+                                        return
+
+                        # Go back to the model request node with an empty request, which means we'll essentially
+                        # resubmit the most recent request that resulted in an empty response,
+                        # as the empty response and request will not create any items in the API payload,
+                        # in the hope the model will return a non-empty response this time.
+                        ctx.state.increment_retries(ctx.deps.max_result_retries)
+                        self._next_node = ModelRequestNode[DepsT, NodeRunEndT](_messages.ModelRequest(parts=[]))
+                except ToolRetryError as e:
+                    ctx.state.increment_retries(ctx.deps.max_result_retries, e)
+                    self._next_node = ModelRequestNode[DepsT, NodeRunEndT](_messages.ModelRequest(parts=[e.tool_retry]))
 
             self._events_iterator = _run_stream()
 
@@ -655,28 +673,22 @@ class CallToolsNode(AgentNode[DepsT, NodeRunEndT]):
     async def _handle_text_response(
         self,
         ctx: GraphRunContext[GraphAgentState, GraphAgentDeps[DepsT, NodeRunEndT]],
-        texts: list[str],
+        text: str,
     ) -> ModelRequestNode[DepsT, NodeRunEndT] | End[result.FinalResult[NodeRunEndT]]:
         output_schema = ctx.deps.output_schema
+        run_context = build_run_context(ctx)
 
-        text = '\n\n'.join(texts)
-        try:
-            run_context = build_run_context(ctx)
-            if isinstance(output_schema, _output.TextOutputSchema):
-                result_data = await output_schema.process(text, run_context)
-            else:
-                m = _messages.RetryPromptPart(
-                    content='Plain text responses are not permitted, please include your response in a tool call',
-                )
-                raise ToolRetryError(m)
-
-            for validator in ctx.deps.output_validators:
-                result_data = await validator.validate(result_data, run_context)
-        except ToolRetryError as e:
-            ctx.state.increment_retries(ctx.deps.max_result_retries, e)
-            return ModelRequestNode[DepsT, NodeRunEndT](_messages.ModelRequest(parts=[e.tool_retry]))
+        if isinstance(output_schema, _output.TextOutputSchema):
+            result_data = await output_schema.process(text, run_context)
         else:
-            return self._handle_final_result(ctx, result.FinalResult(result_data), [])
+            m = _messages.RetryPromptPart(
+                content='Plain text responses are not permitted, please include your response in a tool call',
+            )
+            raise ToolRetryError(m)
+
+        for validator in ctx.deps.output_validators:
+            result_data = await validator.validate(result_data, run_context)
+        return self._handle_final_result(ctx, result.FinalResult(result_data), [])
 
     __repr__ = dataclasses_no_defaults_repr
 

{pydantic_ai_slim-1.0.8 → pydantic_ai_slim-1.0.10}/pydantic_ai/_cli.py

@@ -356,7 +356,7 @@ def handle_slash_command(
         except IndexError:
             console.print('[dim]No output available to copy.[/dim]')
         else:
-            text_to_copy = '\n\n'.join(part.content for part in parts if isinstance(part, TextPart))
+            text_to_copy = ''.join(part.content for part in parts if isinstance(part, TextPart))
             text_to_copy = text_to_copy.strip()
             if text_to_copy:
                 pyperclip.copy(text_to_copy)

{pydantic_ai_slim-1.0.8 → pydantic_ai_slim-1.0.10}/pydantic_ai/_otel_messages.py

@@ -21,6 +21,7 @@ class ToolCallPart(TypedDict):
     id: str
     name: str
     arguments: NotRequired[JsonValue]
+    builtin: NotRequired[bool]  # Not (currently?) part of the spec, used by Logfire
 
 
 class ToolCallResponsePart(TypedDict):
@@ -28,6 +29,7 @@ class ToolCallResponsePart(TypedDict):
     id: str
     name: str
     result: NotRequired[JsonValue]
+    builtin: NotRequired[bool]  # Not (currently?) part of the spec, used by Logfire
 
 
 class MediaUrlPart(TypedDict):

{pydantic_ai_slim-1.0.8 → pydantic_ai_slim-1.0.10}/pydantic_ai/_parts_manager.py

@@ -19,6 +19,8 @@ from typing import Any
 
 from pydantic_ai.exceptions import UnexpectedModelBehavior
 from pydantic_ai.messages import (
+    BuiltinToolCallPart,
+    BuiltinToolReturnPart,
     ModelResponsePart,
     ModelResponseStreamEvent,
     PartDeltaEvent,
@@ -226,11 +228,11 @@ class ModelResponsePartsManager:
         self,
         *,
         vendor_part_id: Hashable | None,
-        tool_name: str | None,
-        args: str | dict[str, Any] | None,
-        tool_call_id: str | None,
+        tool_name: str | None = None,
+        args: str | dict[str, Any] | None = None,
+        tool_call_id: str | None = None,
     ) -> ModelResponseStreamEvent | None:
-        """Handle or update a tool call, creating or updating a `ToolCallPart` or `ToolCallPartDelta`.
+        """Handle or update a tool call, creating or updating a `ToolCallPart`, `BuiltinToolCallPart`, or `ToolCallPartDelta`.
 
         Managed items remain as `ToolCallPartDelta`s until they have at least a tool_name, at which
         point they are upgraded to `ToolCallPart`s.
@@ -247,15 +249,17 @@ class ModelResponsePartsManager:
             tool_call_id: An optional string representing an identifier for this tool call.
 
         Returns:
-            - A `PartStartEvent` if a new ToolCallPart is created.
+            - A `PartStartEvent` if a new ToolCallPart or BuiltinToolCallPart is created.
             - A `PartDeltaEvent` if an existing part is updated.
             - `None` if no new event is emitted (e.g., the part is still incomplete).
 
         Raises:
             UnexpectedModelBehavior: If attempting to apply a tool call delta to a part that is not
-                a ToolCallPart or ToolCallPartDelta.
+                a ToolCallPart, BuiltinToolCallPart, or ToolCallPartDelta.
         """
-        existing_matching_part_and_index: tuple[ToolCallPartDelta | ToolCallPart, int] | None = None
+        existing_matching_part_and_index: tuple[ToolCallPartDelta | ToolCallPart | BuiltinToolCallPart, int] | None = (
+            None
+        )
 
         if vendor_part_id is None:
             # vendor_part_id is None, so check if the latest part is a matching tool call or delta to update
@@ -264,14 +268,14 @@ class ModelResponsePartsManager:
             if tool_name is None and self._parts:
                 part_index = len(self._parts) - 1
                 latest_part = self._parts[part_index]
-                if isinstance(latest_part, ToolCallPart | ToolCallPartDelta):  # pragma: no branch
+                if isinstance(latest_part, ToolCallPart | BuiltinToolCallPart | ToolCallPartDelta):  # pragma: no branch
                     existing_matching_part_and_index = latest_part, part_index
         else:
             # vendor_part_id is provided, so look up the corresponding part or delta
             part_index = self._vendor_id_to_part_index.get(vendor_part_id)
             if part_index is not None:
                 existing_part = self._parts[part_index]
-                if not isinstance(existing_part, ToolCallPartDelta | ToolCallPart):
+                if not isinstance(existing_part, ToolCallPartDelta | ToolCallPart | BuiltinToolCallPart):
                     raise UnexpectedModelBehavior(f'Cannot apply a tool call delta to {existing_part=}')
                 existing_matching_part_and_index = existing_part, part_index
 
@@ -284,7 +288,7 @@ class ModelResponsePartsManager:
             new_part_index = len(self._parts)
             self._parts.append(part)
             # Only emit a PartStartEvent if we have enough information to produce a full ToolCallPart
-            if isinstance(part, ToolCallPart):
+            if isinstance(part, ToolCallPart | BuiltinToolCallPart):
                 return PartStartEvent(index=new_part_index, part=part)
         else:
             # Update the existing part or delta with the new information
@@ -292,7 +296,7 @@ class ModelResponsePartsManager:
             delta = ToolCallPartDelta(tool_name_delta=tool_name, args_delta=args, tool_call_id=tool_call_id)
             updated_part = delta.apply(existing_part)
             self._parts[part_index] = updated_part
-            if isinstance(updated_part, ToolCallPart):
+            if isinstance(updated_part, ToolCallPart | BuiltinToolCallPart):
                 if isinstance(existing_part, ToolCallPartDelta):
                     # We just upgraded a delta to a full part, so emit a PartStartEvent
                     return PartStartEvent(index=part_index, part=updated_part)
@@ -337,7 +341,7 @@ class ModelResponsePartsManager:
         else:
             # vendor_part_id is provided, so find and overwrite or create a new ToolCallPart.
             maybe_part_index = self._vendor_id_to_part_index.get(vendor_part_id)
-            if maybe_part_index is not None:
+            if maybe_part_index is not None and isinstance(self._parts[maybe_part_index], ToolCallPart):
                 new_part_index = maybe_part_index
                 self._parts[new_part_index] = new_part
             else:
@@ -345,3 +349,69 @@ class ModelResponsePartsManager:
                 self._parts.append(new_part)
             self._vendor_id_to_part_index[vendor_part_id] = new_part_index
         return PartStartEvent(index=new_part_index, part=new_part)
+
+    def handle_builtin_tool_call_part(
+        self,
+        *,
+        vendor_part_id: Hashable | None,
+        part: BuiltinToolCallPart,
+    ) -> ModelResponseStreamEvent:
+        """Create or overwrite a BuiltinToolCallPart.
+
+        Args:
+            vendor_part_id: The vendor's ID for this tool call part. If not
+                None and an existing part is found, that part is overwritten.
+            part: The BuiltinToolCallPart.
+
+        Returns:
+            ModelResponseStreamEvent: A `PartStartEvent` indicating that a new tool call part
+            has been added to the manager, or replaced an existing part.
+        """
+        if vendor_part_id is None:
+            # vendor_part_id is None, so we unconditionally append a new BuiltinToolCallPart to the end of the list
+            new_part_index = len(self._parts)
+            self._parts.append(part)
+        else:
+            # vendor_part_id is provided, so find and overwrite or create a new BuiltinToolCallPart.
+            maybe_part_index = self._vendor_id_to_part_index.get(vendor_part_id)
+            if maybe_part_index is not None and isinstance(self._parts[maybe_part_index], BuiltinToolCallPart):
+                new_part_index = maybe_part_index
+                self._parts[new_part_index] = part
+            else:
+                new_part_index = len(self._parts)
+                self._parts.append(part)
+            self._vendor_id_to_part_index[vendor_part_id] = new_part_index
+        return PartStartEvent(index=new_part_index, part=part)
+
+    def handle_builtin_tool_return_part(
+        self,
+        *,
+        vendor_part_id: Hashable | None,
+        part: BuiltinToolReturnPart,
+    ) -> ModelResponseStreamEvent:
+        """Create or overwrite a BuiltinToolReturnPart.
+
+        Args:
+            vendor_part_id: The vendor's ID for this tool call part. If not
+                None and an existing part is found, that part is overwritten.
+            part: The BuiltinToolReturnPart.
+
+        Returns:
+            ModelResponseStreamEvent: A `PartStartEvent` indicating that a new tool call part
+            has been added to the manager, or replaced an existing part.
+        """
+        if vendor_part_id is None:
+            # vendor_part_id is None, so we unconditionally append a new BuiltinToolReturnPart to the end of the list
+            new_part_index = len(self._parts)
+            self._parts.append(part)
+        else:
+            # vendor_part_id is provided, so find and overwrite or create a new BuiltinToolReturnPart.
+            maybe_part_index = self._vendor_id_to_part_index.get(vendor_part_id)
+            if maybe_part_index is not None and isinstance(self._parts[maybe_part_index], BuiltinToolReturnPart):
+                new_part_index = maybe_part_index
+                self._parts[new_part_index] = part
+            else:
+                new_part_index = len(self._parts)
+                self._parts.append(part)
+            self._vendor_id_to_part_index[vendor_part_id] = new_part_index
+        return PartStartEvent(index=new_part_index, part=part)

{pydantic_ai_slim-1.0.8 → pydantic_ai_slim-1.0.10}/pydantic_ai/_run_context.py

@@ -43,10 +43,17 @@ class RunContext(Generic[AgentDepsT]):
     tool_name: str | None = None
     """Name of the tool being called."""
     retry: int = 0
-    """Number of retries so far."""
+    """Number of retries of this tool so far."""
+    max_retries: int = 0
+    """The maximum number of retries of this tool."""
     run_step: int = 0
     """The current step in the run."""
     tool_call_approved: bool = False
     """Whether a tool call that required approval has now been approved."""
 
+    @property
+    def last_attempt(self) -> bool:
+        """Whether this is the last attempt at running this tool before an error is raised."""
+        return self.retry == self.max_retries
+
     __repr__ = _utils.dataclasses_no_defaults_repr

{pydantic_ai_slim-1.0.8 → pydantic_ai_slim-1.0.10}/pydantic_ai/_tool_manager.py

@@ -147,6 +147,7 @@ class ToolManager(Generic[AgentDepsT]):
                 tool_name=name,
                 tool_call_id=call.tool_call_id,
                 retry=self.ctx.retries.get(name, 0),
+                max_retries=tool.max_retries,
             )
 
             pyd_allow_partial = 'trailing-strings' if allow_partial else 'off'

{pydantic_ai_slim-1.0.8 → pydantic_ai_slim-1.0.10}/pydantic_ai/ag_ui.py

@@ -9,7 +9,7 @@ from __future__ import annotations
 import json
 import uuid
 from collections.abc import AsyncIterator, Awaitable, Callable, Iterable, Mapping, Sequence
-from dataclasses import Field, dataclass, replace
+from dataclasses import Field, dataclass, field, replace
 from http import HTTPStatus
 from typing import (
     Any,
@@ -23,13 +23,15 @@ from typing import (
 )
 
 from pydantic import BaseModel, ValidationError
-from typing_extensions import assert_never
 
 from . import _utils
 from ._agent_graph import CallToolsNode, ModelRequestNode
 from .agent import AbstractAgent, AgentRun, AgentRunResult
 from .exceptions import UserError
 from .messages import (
+    BaseToolCallPart,
+    BuiltinToolCallPart,
+    BuiltinToolReturnPart,
     FunctionToolResultEvent,
     ModelMessage,
     ModelRequest,
@@ -123,6 +125,8 @@ SSE_CONTENT_TYPE: Final[str] = 'text/event-stream'
 OnCompleteFunc: TypeAlias = Callable[[AgentRunResult[Any]], None] | Callable[[AgentRunResult[Any]], Awaitable[None]]
 """Callback function type that receives the `AgentRunResult` of the completed run. Can be sync or async."""
 
+_BUILTIN_TOOL_CALL_ID_PREFIX: Final[str] = 'pyd_ai_builtin'
+
 
 class AGUIApp(Generic[AgentDepsT, OutputDataT], Starlette):
     """ASGI application for running Pydantic AI agents with AG-UI protocol support."""
@@ -487,20 +491,37 @@ async def _handle_model_request_event(  # noqa: C901
                 stream_ctx.part_end = TextMessageEndEvent(
                     message_id=message_id,
                 )
-            elif isinstance(part, ToolCallPart):  # pragma: no branch
+            elif isinstance(part, BaseToolCallPart):
+                tool_call_id = part.tool_call_id
+                if isinstance(part, BuiltinToolCallPart):
+                    builtin_tool_call_id = '|'.join(
+                        [_BUILTIN_TOOL_CALL_ID_PREFIX, part.provider_name or '', tool_call_id]
+                    )
+                    stream_ctx.builtin_tool_call_ids[tool_call_id] = builtin_tool_call_id
+                    tool_call_id = builtin_tool_call_id
+
                 message_id = stream_ctx.message_id or stream_ctx.new_message_id()
                 yield ToolCallStartEvent(
-                    tool_call_id=part.tool_call_id,
+                    tool_call_id=tool_call_id,
                     tool_call_name=part.tool_name,
                     parent_message_id=message_id,
                 )
                 if part.args:
                     yield ToolCallArgsEvent(
-                        tool_call_id=part.tool_call_id,
-                        delta=part.args if isinstance(part.args, str) else json.dumps(part.args),
+                        tool_call_id=tool_call_id,
+                        delta=part.args_as_json_str(),
                     )
                 stream_ctx.part_end = ToolCallEndEvent(
-                    tool_call_id=part.tool_call_id,
+                    tool_call_id=tool_call_id,
+                )
+            elif isinstance(part, BuiltinToolReturnPart):  # pragma: no branch
+                tool_call_id = stream_ctx.builtin_tool_call_ids[part.tool_call_id]
+                yield ToolCallResultEvent(
+                    message_id=stream_ctx.new_message_id(),
+                    type=EventType.TOOL_CALL_RESULT,
+                    role='tool',
+                    tool_call_id=tool_call_id,
+                    content=part.model_response_str(),
                 )
 
     elif isinstance(agent_event, PartDeltaEvent):
@@ -512,9 +533,12 @@ async def _handle_model_request_event(  # noqa: C901
                     delta=delta.content_delta,
                 )
         elif isinstance(delta, ToolCallPartDelta):  # pragma: no branch
-            assert delta.tool_call_id, '`ToolCallPartDelta.tool_call_id` must be set'
+            tool_call_id = delta.tool_call_id
+            assert tool_call_id, '`ToolCallPartDelta.tool_call_id` must be set'
+            if tool_call_id in stream_ctx.builtin_tool_call_ids:
+                tool_call_id = stream_ctx.builtin_tool_call_ids[tool_call_id]
             yield ToolCallArgsEvent(
-                tool_call_id=delta.tool_call_id,
+                tool_call_id=tool_call_id,
                 delta=delta.args_delta if isinstance(delta.args_delta, str) else json.dumps(delta.args_delta),
             )
         elif isinstance(delta, ThinkingPartDelta):  # pragma: no branch
@@ -550,9 +574,8 @@ async def _handle_tool_result_event(
     if not isinstance(result, ToolReturnPart):
         return
 
-    message_id = stream_ctx.new_message_id()
     yield ToolCallResultEvent(
-        message_id=message_id,
+        message_id=stream_ctx.new_message_id(),
         type=EventType.TOOL_CALL_RESULT,
         role='tool',
         tool_call_id=result.tool_call_id,
@@ -579,7 +602,9 @@ def _messages_from_ag_ui(messages: list[Message]) -> list[ModelMessage]:
     request_parts: list[ModelRequestPart] | None = None
     response_parts: list[ModelResponsePart] | None = None
     for msg in messages:
-        if isinstance(msg, UserMessage | SystemMessage | DeveloperMessage | ToolMessage):
+        if isinstance(msg, UserMessage | SystemMessage | DeveloperMessage) or (
+            isinstance(msg, ToolMessage) and not msg.tool_call_id.startswith(_BUILTIN_TOOL_CALL_ID_PREFIX)
+        ):
             if request_parts is None:
                 request_parts = []
                 result.append(ModelRequest(parts=request_parts))
@@ -589,44 +614,71 @@ def _messages_from_ag_ui(messages: list[Message]) -> list[ModelMessage]:
                 request_parts.append(UserPromptPart(content=msg.content))
             elif isinstance(msg, SystemMessage | DeveloperMessage):
                 request_parts.append(SystemPromptPart(content=msg.content))
-            elif isinstance(msg, ToolMessage):
-                tool_name = tool_calls.get(msg.tool_call_id)
+            else:
+                tool_call_id = msg.tool_call_id
+                tool_name = tool_calls.get(tool_call_id)
                 if tool_name is None:  # pragma: no cover
-                    raise _ToolCallNotFoundError(tool_call_id=msg.tool_call_id)
+                    raise _ToolCallNotFoundError(tool_call_id=tool_call_id)
 
                 request_parts.append(
                     ToolReturnPart(
                         tool_name=tool_name,
                         content=msg.content,
-                        tool_call_id=msg.tool_call_id,
+                        tool_call_id=tool_call_id,
                     )
                 )
-            else:
-                assert_never(msg)
 
-        elif isinstance(msg, AssistantMessage):
+        elif isinstance(msg, AssistantMessage) or (  # pragma: no branch
+            isinstance(msg, ToolMessage) and msg.tool_call_id.startswith(_BUILTIN_TOOL_CALL_ID_PREFIX)
+        ):
             if response_parts is None:
                 response_parts = []
                 result.append(ModelResponse(parts=response_parts))
                 request_parts = None
 
-            if msg.content:
-                response_parts.append(TextPart(content=msg.content))
-
-            if msg.tool_calls:
-                for tool_call in msg.tool_calls:
-                    tool_calls[tool_call.id] = tool_call.function.name
+            if isinstance(msg, AssistantMessage):
+                if msg.content:
+                    response_parts.append(TextPart(content=msg.content))
+
+                if msg.tool_calls:
+                    for tool_call in msg.tool_calls:
+                        tool_call_id = tool_call.id
+                        tool_name = tool_call.function.name
+                        tool_calls[tool_call_id] = tool_name
+
+                        if tool_call_id.startswith(_BUILTIN_TOOL_CALL_ID_PREFIX):
+                            _, provider_name, tool_call_id = tool_call_id.split('|', 2)
+                            response_parts.append(
+                                BuiltinToolCallPart(
+                                    tool_name=tool_name,
+                                    args=tool_call.function.arguments,
+                                    tool_call_id=tool_call_id,
+                                    provider_name=provider_name,
+                                )
+                            )
+                        else:
+                            response_parts.append(
+                                ToolCallPart(
+                                    tool_name=tool_name,
+                                    tool_call_id=tool_call_id,
+                                    args=tool_call.function.arguments,
+                                )
+                            )
+            else:
+                tool_call_id = msg.tool_call_id
+                tool_name = tool_calls.get(tool_call_id)
+                if tool_name is None:  # pragma: no cover
+                    raise _ToolCallNotFoundError(tool_call_id=tool_call_id)
+                _, provider_name, tool_call_id = tool_call_id.split('|', 2)
 
-                response_parts.extend(
-                    ToolCallPart(
-                        tool_name=tool_call.function.name,
-                        tool_call_id=tool_call.id,
-                        args=tool_call.function.arguments,
+                response_parts.append(
+                    BuiltinToolReturnPart(
+                        tool_name=tool_name,
+                        content=msg.content,
+                        tool_call_id=tool_call_id,
+                        provider_name=provider_name,
                     )
-                    for tool_call in msg.tool_calls
                 )
-        else:
-            assert_never(msg)
 
     return result
 
@@ -687,6 +739,7 @@ class _RequestStreamContext:
     message_id: str = ''
     part_end: BaseEvent | None = None
     thinking: bool = False
+    builtin_tool_call_ids: dict[str, str] = field(default_factory=dict)
 
     def new_message_id(self) -> str:
         """Generate a new message ID for the request stream.

{pydantic_ai_slim-1.0.8 → pydantic_ai_slim-1.0.10}/pydantic_ai/agent/__init__.py

@@ -259,7 +259,8 @@ class Agent(AbstractAgent[AgentDepsT, OutputDataT]):
             name: The name of the agent, used for logging. If `None`, we try to infer the agent name from the call frame
                 when the agent is first run.
             model_settings: Optional model request settings to use for this agent's runs, by default.
-            retries: The default number of retries to allow before raising an error.
+            retries: The default number of retries to allow for tool calls and output validation, before raising an error.
+                For model request retries, see the [HTTP Request Retries](../retries.md) documentation.
             output_retries: The maximum number of retries to allow for output validation, defaults to `retries`.
             tools: Tools to register with the agent, you can also register tools via the decorators
                 [`@agent.tool`][pydantic_ai.Agent.tool] and [`@agent.tool_plain`][pydantic_ai.Agent.tool_plain].