hammad-python 0.0.19__py3-none-any.whl → 0.0.20__py3-none-any.whl

hammad/genai/agents/agent.py

@@ -0,0 +1,1022 @@
+"""hammad.genai.agents.agent"""
+
+from typing import (
+    Any,
+    Callable,
+    Generic,
+    Literal,
+    List,
+    Type,
+    TypeVar,
+    Optional,
+    Union,
+    Dict,
+    overload,
+    TYPE_CHECKING,
+)
+from pydantic import BaseModel, Field, create_model
+from dataclasses import dataclass, field
+from enum import Enum
+import json
+
+from ..types.base import BaseGenAIModel, BaseGenAIModelSettings
+from ..models.language.model import LanguageModel
+from ..models.language.types import (
+    LanguageModelResponse,
+    LanguageModelName,
+    LanguageModelInstructorMode,
+)
+from ..types.tools import (
+    Tool,
+    define_tool,
+    execute_tools_from_language_model_response,
+)
+from ..models.language.utils.requests import (
+    parse_messages_input as parse_messages,
+    consolidate_system_messages,
+)
+from ...formatting.text.converters import convert_to_text
+from .types.agent_response import (
+    AgentResponse,
+    _create_agent_response_from_language_model_response,
+)
+from .types.agent_stream import AgentStream
+from .types.agent_context import AgentContext
+from .types.agent_event import AgentEvent
+from .types.agent_hooks import HookManager, HookDecorator
+from .types.agent_messages import AgentMessages
+
+if TYPE_CHECKING:
+    pass
+
+
+T = TypeVar("T")
+
+
+@dataclass
+class AgentSettings:
+    """Settings object that controls the default behavior of an agent's run."""
+
+    max_steps: int = field(default=10)
+    """The maximum amount of steps the agent can take before stopping."""
+
+    add_name_to_instructions: bool = field(default=True)
+    """Whether to add the agent name to the instructions."""
+
+    context_format: Literal["json", "python", "markdown"] = field(default="json")
+    """Format for context in instructions."""
+
+    # Context management settings
+    context_updates: Optional[
+        Union[List[Literal["before", "after"]], Literal["before", "after"]]
+    ] = field(default=None)
+    """When to update context ('before', 'after', or both)."""
+
+    context_confirm: bool = field(default=False)
+    """Whether to confirm context updates."""
+
+    context_strategy: Literal["selective", "all"] = field(default="all")
+    """Strategy for context updates."""
+
+    context_max_retries: int = field(default=3)
+    """Maximum retries for context updates."""
+
+    context_confirm_instructions: Optional[str] = field(default=None)
+    """Custom instructions for context confirmation."""
+
+    context_selection_instructions: Optional[str] = field(default=None)
+    """Custom instructions for context selection."""
+
+    context_update_instructions: Optional[str] = field(default=None)
+    """Custom instructions for context updates."""
+
+
+class AgentModelSettings(BaseGenAIModelSettings):
+    """Agent-specific model settings that extend the base model settings."""
+
+    instructor_mode: Optional[LanguageModelInstructorMode] = None
+    """Instructor mode for structured outputs."""
+
+    max_steps: int = 10
+    """Maximum number of steps the agent can take."""
+
+    add_name_to_instructions: bool = True
+    """Whether to add the agent name to the instructions."""
+
+    context_format: Literal["json", "python", "markdown"] = "json"
+    """Format for context in instructions."""
+
+    # Context management settings
+    context_updates: Optional[
+        Union[List[Literal["before", "after"]], Literal["before", "after"]]
+    ] = None
+    """When to update context ('before', 'after', or both)."""
+
+    context_confirm: bool = False
+    """Whether to confirm context updates."""
+
+    context_strategy: Literal["selective", "all"] = "all"
+    """Strategy for context updates."""
+
+    context_max_retries: int = 3
+    """Maximum retries for context updates."""
+
+    context_confirm_instructions: Optional[str] = None
+    """Custom instructions for context confirmation."""
+
+    context_selection_instructions: Optional[str] = None
+    """Custom instructions for context selection."""
+
+    context_update_instructions: Optional[str] = None
+    """Custom instructions for context updates."""
+
+
+def _build_tools(tools: List[Tool] | Callable | None) -> List[Tool]:
+    """Builds a list of tools from a list of tools or a callable that returns a list of tools."""
+    if tools is None:
+        return []
+    if callable(tools):
+        return [define_tool(tools)]
+
+    processed_tools = []
+    for tool in tools:
+        if not isinstance(tool, Tool):
+            tool = define_tool(tool)
+        processed_tools.append(tool)
+
+    return processed_tools
+
+
+def _get_instructions(
+    name: str,
+    instructions: Optional[str],
+    add_name_to_instructions: bool,
+) -> Optional[str]:
+    """Gets the instructions for an agent."""
+    if add_name_to_instructions and name:
+        base_instructions = instructions or ""
+        return f"You are {name}.\n\n{base_instructions}".strip()
+    return instructions
+
+
+def _format_context_for_instructions(
+    context: AgentContext | None,
+    context_format: Literal["json", "python", "markdown"] = "json",
+) -> str:
+    """Format context object for inclusion in instructions."""
+    if context is None:
+        return ""
+
+    if context_format == "json":
+        if isinstance(context, BaseModel):
+            return context.model_dump_json(indent=2)
+        elif isinstance(context, dict):
+            return json.dumps(context, indent=2)
+        else:
+            return json.dumps(str(context), indent=2)
+
+    elif context_format == "python":
+        if hasattr(context, "__repr__"):
+            return repr(context)
+        elif hasattr(context, "__str__"):
+            return str(context)
+        else:
+            return str(context)
+
+    elif context_format == "markdown":
+        return convert_to_text(context)
+
+    return str(context)
+
+
+def _update_context_object(
+    context: AgentContext, updates: Dict[str, Any]
+) -> AgentContext:
+    """Update a context object with new values."""
+    if isinstance(context, BaseModel):
+        # For Pydantic models, create a copy with updated values
+        return context.model_copy(update=updates)
+    elif isinstance(context, dict):
+        # For dictionaries, update in place
+        updated_context = context.copy()
+        updated_context.update(updates)
+        return updated_context
+    else:
+        raise ValueError(f"Cannot update context of type {type(context)}")
+
+
+class Agent(BaseGenAIModel, Generic[T]):
+    """A generative AI agent that can execute tools, generate structured outputs,
+    and maintain context across multiple conversation steps.
+    """
+
+    model: LanguageModelName = "openai/gpt-4o-mini"
+    """The language model to use for the agent."""
+
+    name: str = "agent"
+    """The name of the agent."""
+
+    description: Optional[str] = None
+    """A description of the agent."""
+
+    instructions: Optional[str] = None
+    """System instructions for the agent."""
+
+    tools: List[Tool] = Field(default_factory=list)
+    """List of tools available to the agent."""
+
+    settings: AgentSettings = Field(default_factory=AgentSettings)
+    """Agent-specific settings."""
+
+    instructor_mode: Optional[LanguageModelInstructorMode] = None
+    """Instructor mode for structured outputs."""
+
+    def __init__(
+        self,
+        name: str = "agent",
+        instructions: Optional[str] = None,
+        model: Union[LanguageModel, LanguageModelName] = "openai/gpt-4o-mini",
+        description: Optional[str] = None,
+        tools: Union[List[Tool], Callable, None] = None,
+        settings: Optional[AgentSettings] = None,
+        instructor_mode: Optional[LanguageModelInstructorMode] = None,
+        # Context management parameters
+        context_updates: Optional[
+            Union[List[Literal["before", "after"]], Literal["before", "after"]]
+        ] = None,
+        context_confirm: bool = False,
+        context_strategy: Literal["selective", "all"] = "all",
+        context_max_retries: int = 3,
+        context_confirm_instructions: Optional[str] = None,
+        context_selection_instructions: Optional[str] = None,
+        context_update_instructions: Optional[str] = None,
+        context_format: Literal["json", "python", "markdown"] = "json",
+        **kwargs: Any,
+    ):
+        # Initialize BaseGenAIModel with basic parameters
+        super().__init__(
+            model=model if isinstance(model, str) else model.model, **kwargs
+        )
+
+        # Agent-specific initialization
+        self.name = name
+        self.description = description
+        self.tools = _build_tools(tools)
+        self.settings = settings or AgentSettings()
+        self.instructor_mode = instructor_mode
+
+        # Process instructions
+        self.instructions = _get_instructions(
+            name=name,
+            instructions=instructions,
+            add_name_to_instructions=self.settings.add_name_to_instructions,
+        )
+
+        # Initialize the language model
+        if isinstance(model, LanguageModel):
+            self._language_model = model
+        else:
+            self._language_model = LanguageModel(model=model, **kwargs)
+
+        # Context management settings
+        self.context_updates = context_updates
+        self.context_confirm = context_confirm
+        self.context_strategy = context_strategy
+        self.context_max_retries = context_max_retries
+        self.context_confirm_instructions = context_confirm_instructions
+        self.context_selection_instructions = context_selection_instructions
+        self.context_update_instructions = context_update_instructions
+        self.context_format = context_format
+
+        # Hook system
+        self.hook_manager = HookManager()
+        self.on = HookDecorator(self.hook_manager)
+
+    @property
+    def language_model(self) -> LanguageModel:
+        """Get the underlying language model."""
+        return self._language_model
+
+    def _should_update_context(
+        self, context: AgentContext, timing: Literal["before", "after"]
+    ) -> bool:
+        """Determine if context should be updated based on timing and configuration."""
+        if not self.context_updates:
+            return False
+
+        if isinstance(self.context_updates, str):
+            return self.context_updates == timing
+        else:
+            return timing in self.context_updates
+
+    def _create_context_confirm_model(self):
+        """Create IsUpdateRequired model for context confirmation."""
+        return create_model("IsUpdateRequired", decision=(bool, ...))
+
+    def _create_context_selection_model(self, context: AgentContext):
+        """Create FieldsToUpdate model for selective context updates."""
+        if isinstance(context, BaseModel):
+            field_names = list(context.model_fields.keys())
+        elif isinstance(context, dict):
+            field_names = list(context.keys())
+        else:
+            raise ValueError(
+                f"Cannot create selection model for context type {type(context)}"
+            )
+
+        FieldEnum = Enum("FieldEnum", {name: name for name in field_names})
+        return create_model("FieldsToUpdate", fields=(List[FieldEnum], ...))
+
+    def _create_context_update_model(
+        self, context: AgentContext, field_name: str = None
+    ):
+        """Create update model for context updates."""
+        if field_name:
+            # Single field update
+            if isinstance(context, BaseModel):
+                field_type = context.model_fields[field_name].annotation
+            elif isinstance(context, dict):
+                field_type = type(context[field_name])
+            else:
+                field_type = Any
+
+            return create_model(
+                field_name.capitalize(), **{field_name: (field_type, ...)}
+            )
+        else:
+            # All fields update
+            return create_model("Update", updates=(Dict[str, Any], ...))
+
+    def _perform_context_update(
+        self,
+        context: AgentContext,
+        model: LanguageModel,
+        current_messages: List[Dict[str, Any]],
+        timing: Literal["before", "after"],
+    ) -> AgentContext:
+        """Perform context update with retries and error handling."""
+        updated_context = context
+
+        for attempt in range(self.context_max_retries):
+            try:
+                # Check if update is needed (if confirmation is enabled)
+                if self.context_confirm:
+                    confirm_model = self._create_context_confirm_model()
+                    confirm_instructions = f"Based on the conversation, determine if the context should be updated {timing} processing."
+                    if self.context_confirm_instructions:
+                        confirm_instructions += (
+                            f"\n\n{self.context_confirm_instructions}"
+                        )
+
+                    confirm_response = model.run(
+                        messages=current_messages
+                        + [{"role": "user", "content": confirm_instructions}],
+                        type=confirm_model,
+                        instructor_mode=self.instructor_mode,
+                    )
+
+                    if not confirm_response.output.decision:
+                        return updated_context
+
+                # Perform the update based on strategy
+                if self.context_strategy == "selective":
+                    # Get fields to update
+                    selection_model = self._create_context_selection_model(
+                        updated_context
+                    )
+                    selection_instructions = f"Select which fields in the context should be updated {timing} processing."
+                    if self.context_selection_instructions:
+                        selection_instructions += (
+                            f"\n\n{self.context_selection_instructions}"
+                        )
+
+                    selection_response = model.run(
+                        messages=current_messages
+                        + [{"role": "user", "content": selection_instructions}],
+                        type=selection_model,
+                        instructor_mode=self.instructor_mode,
+                    )
+
+                    # Update each selected field
+                    for field_enum in selection_response.output.fields:
+                        field_name = field_enum.value
+                        field_model = self._create_context_update_model(
+                            updated_context, field_name
+                        )
+                        field_instructions = (
+                            f"Update the {field_name} field in the context."
+                        )
+                        if self.context_update_instructions:
+                            field_instructions += (
+                                f"\n\n{self.context_update_instructions}"
+                            )
+
+                        field_response = model.run(
+                            messages=current_messages
+                            + [{"role": "user", "content": field_instructions}],
+                            type=field_model,
+                            instructor_mode=self.instructor_mode,
+                        )
+
+                        # Apply the update
+                        field_updates = {
+                            field_name: getattr(field_response.output, field_name)
+                        }
+                        updated_context = _update_context_object(
+                            updated_context, field_updates
+                        )
+
+                else:  # strategy == "all"
+                    # Update all fields at once
+                    update_model = self._create_context_update_model(updated_context)
+                    update_instructions = f"Update the context {timing} processing."
+                    if self.context_update_instructions:
+                        update_instructions += f"\n\n{self.context_update_instructions}"
+
+                    update_response = model.run(
+                        messages=current_messages
+                        + [{"role": "user", "content": update_instructions}],
+                        type=update_model,
+                        instructor_mode=self.instructor_mode,
+                    )
+
+                    # Apply the updates
+                    updated_context = _update_context_object(
+                        updated_context, update_response.output.updates
+                    )
+
+                # Trigger context update hooks
+                self.hook_manager.trigger_hooks("context_update", updated_context)
+
+                return updated_context
+
+            except Exception as e:
+                if attempt == self.context_max_retries - 1:
+                    # Last attempt failed, return original context
+                    return updated_context
+                # Continue to next attempt
+                continue
+
+        return updated_context
+
+    def _format_messages_with_context(
+        self, messages: List[Dict[str, Any]], context: Optional[AgentContext] = None
+    ) -> List[Dict[str, Any]]:
+        """Format messages with instructions and context."""
+        formatted_messages = messages.copy()
+
+        if self.instructions:
+            system_content = self.instructions
+
+            # Add context if provided
+            if context is not None:
+                context_str = _format_context_for_instructions(
+                    context, self.context_format
+                )
+                if context_str:
+                    system_content += f"\n\nContext:\n{context_str}"
+
+            system_message = {"role": "system", "content": system_content}
+            formatted_messages = [system_message] + formatted_messages
+
+        return consolidate_system_messages(formatted_messages)
+
+    # Overloaded run methods for streaming support
+    @overload
+    def run(
+        self,
+        messages: AgentMessages,
+        model: Optional[Union[LanguageModel, LanguageModelName]] = None,
+        max_steps: Optional[int] = None,
+        context: Optional[AgentContext] = None,
+        output_type: Optional[Type[T]] = None,
+        *,
+        stream: Literal[False] = False,
+        **kwargs: Any,
+    ) -> AgentResponse[T]: ...
+
+    @overload
+    def run(
+        self,
+        messages: AgentMessages,
+        model: Optional[Union[LanguageModel, LanguageModelName]] = None,
+        max_steps: Optional[int] = None,
+        context: Optional[AgentContext] = None,
+        output_type: Optional[Type[T]] = None,
+        *,
+        stream: Literal[True],
+        **kwargs: Any,
+    ) -> AgentStream[T]: ...
+
+    def run(
+        self,
+        messages: AgentMessages,
+        model: Optional[Union[LanguageModel, LanguageModelName]] = None,
+        max_steps: Optional[int] = None,
+        context: Optional[AgentContext] = None,
+        output_type: Optional[Type[T]] = None,
+        stream: bool = False,
+        **kwargs: Any,
+    ) -> Union[AgentResponse[T], AgentStream[T]]:
+        """Runs this agent and returns a final agent response or stream.
+
+        You can override defaults assigned to this agent from this function directly.
+
+        Args:
+            messages: The messages to process. Can be:
+                - A single string: "What's the weather like?"
+                - A list of message dicts: [{"role": "user", "content": "Hello"}]
+                - A list of strings: ["Hello", "How are you?"]
+            model: The model to use for this run (overrides default).
+                - Can be a LanguageModel instance or model name string like "gpt-4"
+            max_steps: Maximum number of steps to execute (overrides default).
+                - Useful for limiting tool usage or preventing infinite loops
+            context: Context object for the agent (overrides default).
+                - Any object that provides additional context for the conversation
+            output_type: The expected output type (overrides default).
+                - Use for structured outputs: output_type=MyPydanticModel
+                - Defaults to str for unstructured text responses
+            stream: Whether to return a stream instead of a final response.
+                - If True, returns AgentStream for real-time processing
+                - If False, returns complete AgentResponse
+            **kwargs: Additional keyword arguments passed to the language model.
+                - Examples: temperature=0.7, top_p=0.9, presence_penalty=0.1
+
+        Returns:
+            AgentResponse or AgentStream depending on stream parameter.
+            - AgentResponse: Contains final output, steps taken, and metadata
+            - AgentStream: Iterator yielding intermediate steps and final result
+
+        Examples:
+            Basic text conversation:
+            >>> agent = Agent()
+            >>> response = agent.run("Hello, how are you?")
+            >>> print(response.output)
+            "Hello! I'm doing well, thank you for asking."
+
+            With custom model and parameters:
+            >>> response = agent.run(
+            ...     messages="Explain quantum computing",
+            ...     model="gpt-4",
+            ...     max_steps=5,
+            ...     temperature=0.3
+            ... )
+
+            Structured output with Pydantic model:
+            >>> from pydantic import BaseModel
+            >>> class Summary(BaseModel):
+            ...     title: str
+            ...     key_points: List[str]
+            >>> response = agent.run(
+            ...     "Summarize the benefits of renewable energy",
+            ...     output_type=Summary
+            ... )
+            >>> print(response.output.title)
+            >>> print(response.output.key_points)
+
+            Streaming for real-time results:
+            >>> stream = agent.run(
+            ...     "Write a long story about space exploration",
+            ...     stream=True
+            ... )
+            >>> for chunk in stream:
+            ...     print(chunk.output, end="", flush=True)
+
+            With context for additional information:
+            >>> context = {"user_preferences": "technical explanations"}
+            >>> response = agent.run(
+            ...     "How does machine learning work?",
+            ...     context=context
+            ... )
+        """
+        # Handle streaming
+        if stream:
+            return AgentStream(
+                agent=self,
+                messages=messages,
+                model=model,
+                max_steps=max_steps,
+                context=context,
+                output_type=output_type,
+                stream=stream,
+                **kwargs,
+            )
+
+        # Use provided model or default
+        if model is None:
+            working_model = self.language_model
+        elif isinstance(model, str):
+            working_model = LanguageModel(model=model)
+        else:
+            working_model = model
+
+        # Use provided max_steps or default
+        if max_steps is None:
+            max_steps = self.settings.max_steps
+
+        # Parse initial messages
+        parsed_messages = parse_messages(messages)
+        current_messages = parsed_messages.copy()
+        steps: List[LanguageModelResponse[str]] = []
+
+        # RUN MAIN AGENTIC LOOP
+        for step in range(max_steps):
+            # Format messages with instructions and context for first step only
+            if step == 0:
+                formatted_messages = self._format_messages_with_context(
+                    messages=current_messages,
+                    context=context,
+                )
+            else:
+                formatted_messages = current_messages
+
+            # Prepare kwargs for language model
+            model_kwargs = kwargs.copy()
+            if output_type:
+                model_kwargs["type"] = output_type
+            if self.instructor_mode:
+                model_kwargs["instructor_mode"] = self.instructor_mode
+
+            # Get language model response
+            response = working_model.run(
+                messages=formatted_messages,
+                tools=[tool.model_dump() for tool in self.tools]
+                if self.tools
+                else None,
+                **model_kwargs,
+            )
+
+            # Check if response has tool calls
+            if response.has_tool_calls():
+                # Add response to message history (with tool calls)
+                current_messages.append(response.to_message())
+
+                # Execute tools and add their responses to messages
+                tool_responses = execute_tools_from_language_model_response(
+                    tools=self.tools, response=response
+                )
+                # Add tool responses to message history
+                for tool_resp in tool_responses:
+                    current_messages.append(tool_resp.to_dict())
+
+                # This is not the final step, add to steps
+                steps.append(response)
+            else:
+                # No tool calls - this is the final step
+                return _create_agent_response_from_language_model_response(
+                    response=response, steps=steps, context=context
+                )
+
+        # Max steps reached - return last response
+        if steps:
+            final_response = steps[-1]
+        else:
+            # No steps taken, make a final call
+            final_response = working_model.run(
+                messages=self._format_messages_with_context(
+                    messages=current_messages,
+                    context=context,
+                ),
+                **model_kwargs,
+            )
+
+        return _create_agent_response_from_language_model_response(
+            response=final_response, steps=steps, context=context
+        )
+
+    async def async_run(
+        self,
+        messages: AgentMessages,
+        model: Optional[Union[LanguageModel, LanguageModelName]] = None,
+        max_steps: Optional[int] = None,
+        context: Optional[AgentContext] = None,
+        output_type: Optional[Type[T]] = None,
+        **kwargs: Any,
+    ) -> AgentResponse[T]:
+        """Runs this agent asynchronously and returns a final agent response.
+
+        You can override defaults assigned to this agent from this function directly.
+        This is the async version of run() for non-blocking execution.
+
+        Args:
+            messages: The messages to process. Can be:
+                - A single string: "What's the weather like?"
+                - A list of message dicts: [{"role": "user", "content": "Hello"}]
+                - A list of strings: ["Hello", "How are you?"]
+            model: The model to use for this run (overrides default).
+                - Can be a LanguageModel instance or model name string like "gpt-4"
+            max_steps: Maximum number of steps to execute (overrides default).
+                - Useful for limiting tool usage or preventing infinite loops
+            context: Context object for the agent (overrides default).
+                - Any object that provides additional context for the conversation
+            output_type: The expected output type (overrides default).
+                - Use for structured outputs: output_type=MyPydanticModel
+                - Defaults to str for unstructured text responses
+            **kwargs: Additional keyword arguments passed to the language model.
+                - Examples: temperature=0.7, top_p=0.9, presence_penalty=0.1
+
+        Returns:
+            AgentResponse containing the final output, steps taken, and metadata.
+
+        Examples:
+            Basic async usage:
+            >>> import asyncio
+            >>> agent = Agent()
+            >>> async def main():
+            ...     response = await agent.async_run("Hello, how are you?")
+            ...     print(response.output)
+            >>> asyncio.run(main())
+
+            Multiple concurrent requests:
+            >>> async def process_multiple():
+            ...     tasks = [
+            ...         agent.async_run("What's 2+2?"),
+            ...         agent.async_run("What's the capital of France?"),
+            ...         agent.async_run("Explain photosynthesis")
+            ...     ]
+            ...     responses = await asyncio.gather(*tasks)
+            ...     return responses
+
+            With structured output:
+            >>> from pydantic import BaseModel
+            >>> class Analysis(BaseModel):
+            ...     sentiment: str
+            ...     confidence: float
+            >>> async def analyze_text():
+            ...     response = await agent.async_run(
+            ...         "Analyze the sentiment of: 'I love this product!'",
+            ...         output_type=Analysis
+            ...     )
+            ...     return response.output
+
+            With custom model and context:
+            >>> async def custom_run():
+            ...     context = {"domain": "medical", "expertise_level": "expert"}
+            ...     response = await agent.async_run(
+            ...         "Explain diabetes",
+            ...         model="gpt-4",
+            ...         context=context,
+            ...         temperature=0.2
+            ...     )
+            ...     return response.output
+        """
+        # Use provided model or default
+        if model is None:
+            working_model = self.language_model
+        elif isinstance(model, str):
+            working_model = LanguageModel(model=model)
+        else:
+            working_model = model
+
+        # Use provided max_steps or default
+        if max_steps is None:
+            max_steps = self.settings.max_steps
+
+        # Parse initial messages
+        parsed_messages = parse_messages(messages)
+        current_messages = parsed_messages.copy()
+        steps: List[LanguageModelResponse[str]] = []
+
+        # RUN MAIN AGENTIC LOOP
+        for step in range(max_steps):
+            # Format messages with instructions and context for first step only
+            if step == 0:
+                formatted_messages = self._format_messages_with_context(
+                    messages=current_messages,
+                    context=context,
+                )
+            else:
+                formatted_messages = current_messages
+
+            # Prepare kwargs for language model
+            model_kwargs = kwargs.copy()
+            if output_type:
+                model_kwargs["type"] = output_type
+            if self.instructor_mode:
+                model_kwargs["instructor_mode"] = self.instructor_mode
+
+            # Get language model response
+            response = await working_model.async_run(
+                messages=formatted_messages,
+                tools=[tool.model_dump() for tool in self.tools]
+                if self.tools
+                else None,
+                **model_kwargs,
+            )
+
+            # Check if response has tool calls
+            if response.has_tool_calls():
+                # Add response to message history (with tool calls)
+                current_messages.append(response.to_message())
+
+                # Execute tools and add their responses to messages
+                tool_responses = execute_tools_from_language_model_response(
+                    tools=self.tools, response=response
+                )
+                # Add tool responses to message history
+                for tool_resp in tool_responses:
+                    current_messages.append(tool_resp.to_dict())
+
+                # This is not the final step, add to steps
+                steps.append(response)
+            else:
+                # No tool calls - this is the final step
+                return _create_agent_response_from_language_model_response(
+                    response=response, steps=steps, context=context
+                )
+
+        # Max steps reached - return last response
+        if steps:
+            final_response = steps[-1]
+        else:
+            # No steps taken, make a final call
+            final_response = await working_model.async_run(
+                messages=self._format_messages_with_context(
+                    messages=current_messages,
+                    context=context,
+                ),
+                **model_kwargs,
+            )
+
+        return _create_agent_response_from_language_model_response(
+            response=final_response, steps=steps, context=context
+        )
+
+    def stream(
+        self,
+        messages: AgentMessages,
+        model: Optional[Union[LanguageModel, LanguageModelName]] = None,
+        max_steps: Optional[int] = None,
+        context: Optional[AgentContext] = None,
+        output_type: Optional[Type[T]] = None,
+        **kwargs: Any,
+    ) -> AgentStream[T]:
+        """Create a stream that yields agent steps.
+
+        Args:
+            messages: The input messages to process
+            model: Language model to use (overrides agent's default)
+            max_steps: Maximum number of steps to take
+            context: Context object to maintain state
+            output_type: Type for structured output
+            **kwargs: Additional parameters for the language model
+
+        Returns:
+            An AgentStream that can be iterated over
+        """
+        return AgentStream(
+            agent=self,
+            messages=messages,
+            model=model,
+            max_steps=max_steps,
+            context=context,
+            output_type=output_type,
+            stream=True,
+            **kwargs,
+        )
+
+    def iter(
+        self,
+        messages: AgentMessages,
+        model: Optional[Union[LanguageModel, LanguageModelName]] = None,
+        max_steps: Optional[int] = None,
+        context: Optional[AgentContext] = None,
+        output_type: Optional[Type[T]] = None,
+        **kwargs: Any,
+    ) -> AgentStream[T]:
+        """Iterate over agent steps, yielding each step response.
+
+        You can override defaults assigned to this agent from this function directly.
+        Returns an AgentStream that yields intermediate steps and the final result.
+
+        Args:
+            messages: The messages to process. Can be:
+                - A single string: "What's the weather like?"
+                - A list of message dicts: [{"role": "user", "content": "Hello"}]
+                - A list of strings: ["Hello", "How are you?"]
+            model: The model to use for this run (overrides default).
+                - Can be a LanguageModel instance or model name string like "gpt-4"
+            max_steps: Maximum number of steps to execute (overrides default).
+                - Useful for limiting tool usage or preventing infinite loops
+            context: Context object for the agent (overrides default).
+                - Any object that provides additional context for the conversation
+            output_type: The expected output type (overrides default).
+                - Use for structured outputs: output_type=MyPydanticModel
+                - Defaults to str for unstructured text responses
+            **kwargs: Additional keyword arguments passed to the language model.
+                - Examples: temperature=0.7, top_p=0.9, presence_penalty=0.1
+
+        Returns:
+            AgentStream that can be iterated over to get each step response,
+            including tool calls and intermediate reasoning steps.
+
+        Examples:
+            Basic iteration over steps:
+            >>> agent = Agent(tools=[calculator_tool])
+            >>> stream = agent.iter("What's 25 * 47?")
+            >>> for step in stream:
+            ...     print(f"Step {step.step_number}: {step.output}")
+            ...     if step.tool_calls:
+            ...         print(f"Tool calls: {len(step.tool_calls)}")
+
+            Real-time processing with streaming:
+            >>> stream = agent.iter("Write a poem about nature")
+            >>> for chunk in stream:
+            ...     if chunk.output:
+            ...         print(chunk.output, end="", flush=True)
+            ...     if chunk.is_final:
+            ...         print("\n--- Final response ---")
+
+            With structured output iteration:
+            >>> from pydantic import BaseModel
+            >>> class StepAnalysis(BaseModel):
+            ...     reasoning: str
+            ...     confidence: float
+            >>> stream = agent.iter(
+            ...     "Analyze this step by step: Why is the sky blue?",
+            ...     output_type=StepAnalysis
+            ... )
+            >>> for step in stream:
+            ...     if step.output:
+            ...         print(f"Reasoning: {step.output.reasoning}")
+            ...         print(f"Confidence: {step.output.confidence}")
+
+            Processing with custom model and context:
+            >>> context = {"domain": "science", "depth": "detailed"}
+            >>> stream = agent.iter(
+            ...     "Explain quantum entanglement",
+            ...     model="gpt-4",
+            ...     context=context,
+            ...     max_steps=3,
+            ...     temperature=0.1
+            ... )
+            >>> results = []
+            >>> for step in stream:
+            ...     results.append(step.output)
+            ...     if step.is_final:
+            ...         break
+
+            Error handling during iteration:
+            >>> try:
+            ...     stream = agent.iter("Complex calculation task")
+            ...     for step in stream:
+            ...         if step.error:
+            ...             print(f"Error in step: {step.error}")
+            ...         else:
+            ...             print(f"Step result: {step.output}")
+            ... except Exception as e:
+            ...     print(f"Stream error: {e}")
+        """
+        return AgentStream(
+            agent=self,
+            messages=messages,
+            model=model,
+            max_steps=max_steps,
+            context=context,
+            output_type=output_type,
+            stream=True,
+            **kwargs,
+        )
+
+    def async_iter(
+        self,
+        messages: AgentMessages,
+        model: Optional[Union[LanguageModel, LanguageModelName]] = None,
+        max_steps: Optional[int] = None,
+        context: Optional[AgentContext] = None,
+        output_type: Optional[Type[T]] = None,
+        **kwargs: Any,
+    ) -> AgentStream[T]:
+        """Async iterate over agent steps, yielding each step response.
+
+        Args:
+            messages: The input messages to process
+            model: Language model to use (overrides agent's default)
+            max_steps: Maximum number of steps to take
+            context: Context object to maintain state
+            output_type: Type for structured output
+            **kwargs: Additional parameters for the language model
+
+        Returns:
+            An AgentStream that can be iterated over asynchronously
+        """
+        return AgentStream(
+            agent=self,
+            messages=messages,
+            model=model,
+            max_steps=max_steps,
+            context=context,
+            output_type=output_type,
+            stream=True,
+            **kwargs,
+        )
+
+
+__all__ = [
+    "Agent",
+    "AgentSettings",
+    "AgentModelSettings",
+    "AgentEvent",
+    "HookManager",
+    "HookDecorator",
+]