flock-core 0.1.2__py3-none-any.whl → 0.2.1__py3-none-any.whl

flock/core/flock_agent.py

@@ -0,0 +1,492 @@
+"""FlockAgent is the core, declarative base class for all agents in the Flock framework."""
+
+from abc import ABC
+from collections.abc import Callable
+from dataclasses import dataclass, field
+from typing import Any, Literal, Union
+
+import cloudpickle
+from pydantic import BaseModel, Field
+
+from flock.core.context.context import FlockContext
+from flock.core.logging.logging import get_logger
+from flock.core.mixin.dspy_integration import DSPyIntegrationMixin
+from flock.core.mixin.prompt_parser import PromptParserMixin
+
+logger = get_logger("flock")
+
+
+@dataclass
+class FlockAgentConfig:
+    """Configuration options for a FlockAgent."""
+
+    save_to_file: bool = field(
+        default=False,
+        metadata={
+            "description": "Saves the serialized agent to a file every time it gets serialized."
+        },
+    )
+    data_type: Literal["json", "cloudpickle", "msgpack"] = "cloudpickle"
+
+
+@dataclass
+class HandoffBase:
+    """Base class for handoff returns."""
+
+    next_agent: Union[str, "FlockAgent"] = field(
+        default="", metadata={"description": "Next agent to invoke"}
+    )
+    input: dict[str, Any] = field(
+        default_factory=dict,
+        metadata={"description": "Input data for the next agent"},
+    )
+    context: FlockContext = field(
+        default=None, metadata={"description": "Override context parameters"}
+    )
+
+
+class FlockAgent(BaseModel, ABC, PromptParserMixin, DSPyIntegrationMixin):
+    """FlockAgent is the core, declarative base class for all agents in the Flock framework.
+
+    Due to its declarative nature, FlockAgent does not rely on constructing classical prompts manually.
+    Instead, each agent is defined by simply declaring:
+    - Its expected inputs (what data it needs),
+    - Its expected outputs (what data it produces), and
+    - Any optional tools it can use during execution.
+
+    In a declarative model, you describe *what* you expect rather than *how* to compute it. This means that
+    instead of embedding prompt engineering logic directly in your code, you specify a concise signature.
+    At runtime, the Flock framework automatically:
+    - Resolves the inputs from a pre-computed context,
+    - Constructs a precise prompt for the underlying language model (using metadata such as type hints
+        and human-readable descriptions), and
+    - Invokes the appropriate tools if needed.
+
+    This approach minimizes hidden dependencies and boilerplate code. It allows developers to focus solely on
+    what data the agent should work with and what result it should produce, without worrying about the intricacies
+    of prompt formatting or context management.
+
+    For details on how Flock resolves inputs, please refer to the documentation.
+
+    Key benefits of the declarative approach include:
+    - **Clarity:** The agent's interface (inputs and outputs) is explicitly defined.
+    - **Reusability:** Agents can be easily serialized, shared, and reused since they are built as Pydantic models.
+    - **Modularity:** By receiving a dictionary of pre-resolved inputs, agents operate independently of the global context,
+        making them easier to test and debug.
+    - **Extensibility:** Additional metadata (like detailed descriptions for each key) can be embedded and later used to
+        refine the prompt for the LLM.
+
+    Since FlockAgent is a Pydantic BaseModel, it can be serialized to and from JSON. This ensures that the agent's
+    configuration and state are easily stored, transmitted, and reproduced.
+
+    **Implementation Example:**
+
+    Below is an example of how to define and instantiate a FlockAgent:
+
+        from flock_agent import FlockAgent
+        import basic_tools
+
+        # Define an agent by declaring its inputs, outputs, and optional tools.
+        idea_agent = FlockAgent(
+            name="idea_agent",
+            input="query: str | The search query, context: dict | The full conversation context",
+            output="a_fun_software_project_idea: str | The generated software project idea",
+            tools=[basic_tools.web_search_tavily],
+        )
+
+        # At runtime, Flock automatically resolves inputs and calls the agent:
+        resolved_inputs = {
+            "query": "A new social media app",
+            "context": {"previous_idea": "a messaging platform"}
+        }
+        result = await idea_agent.run(resolved_inputs)
+
+    In this example:
+    - The agent declares that it needs a `query` (a string describing what to search for) and a `context` (a dictionary
+        containing additional information).
+    - It produces a `a_fun_software_project_idea` (a string with the generated idea).
+    - The tool `basic_tools.web_search_tavily` is available for the agent to use if needed.
+    - When the agent is run, the Flock framework resolves the inputs, constructs the appropriate prompt using the
+        declared metadata, and returns the result.
+
+    This declarative style streamlines agent creation and execution, making the framework highly modular, testable,
+    and production-ready.
+
+    In the future options will be provided to optimize the "hidden" prompt generation and execution so the agent will
+    perform close to its theoretical maximum.
+    """
+
+    name: str = Field(..., description="Unique identifier for the agent.")
+    model: str = Field(
+        "openai/gpt-4o", description="The model to use (e.g., 'openai/gpt-4o')."
+    )
+    description: str | Callable[..., str] = Field(
+        "", description="A human-readable description of the agent."
+    )
+
+    input: str | None = Field(
+        None,
+        description=(
+            "A comma-separated list of input keys. Optionally supports type hints (:) and descriptions (|). "
+            "For example: 'query: str | The search query, chapter_list: list[str] | The chapter list of the document'."
+        ),
+    )
+    output: str | None = Field(
+        None,
+        description=(
+            "A comma-separated list of output keys.  Optionally supports type hints (:) and descriptions (|). "
+            "For example: 'result|The generated result, summary|A brief summary'."
+        ),
+    )
+
+    tools: list[Callable[..., Any]] | None = Field(
+        default=None,
+        description="An optional list of callable tools that the agent can leverage during execution.",
+    )
+
+    use_cache: bool = Field(
+        default=False,
+        description="Set to True to enable caching of the agent's results.",
+    )
+
+    hand_off: str | Callable[..., Any] | None = Field(
+        None,
+        description=(
+            "Specifies the next agent in the workflow or a callable that determines the handoff. "
+            "This allows chaining of agents."
+        ),
+    )
+
+    termination: str | None = Field(
+        None,
+        description="An optional termination condition or phrase used to indicate when the agent should stop processing.",
+    )
+
+    config: FlockAgentConfig = Field(
+        default_factory=FlockAgentConfig,
+        description="Configuration options for the agent, such as serialization settings.",
+    )
+
+    # Lifecycle hooks
+    async def initialize(self, inputs: dict[str, Any]) -> None:
+        """The very first thing to get called.
+
+        Override this method to perform any setup or configuration tasks,
+        such as loading resources or validating inputs.
+        """
+        pass
+
+    async def terminate(
+        self, inputs: dict[str, Any], result: dict[str, Any]
+    ) -> None:
+        """The very last thing to get called.
+
+        Override this method to perform any cleanup tasks,
+        such as releasing resources or logging results.
+        """
+        pass
+
+    async def on_error(self, error: Exception, inputs: dict[str, Any]) -> None:
+        """Called if the agent encounters an error during execution.
+
+        Override this method to implement
+        custom error handling or recovery strategies.
+        """
+        pass
+
+    async def evaluate(self, inputs: dict[str, Any]) -> dict[str, Any]:
+        """Process the agent's task using the provided inputs and return the result.
+
+        This asynchronous method is the core execution engine for a FlockAgent. It performs the following steps:
+
+        1. **Extract Descriptions:**
+            Parses the agent's configured input and output strings to extract human-readable descriptions.
+            These strings are expected to use the format "key: type_hint | description". The method removes the
+            type hints and builds dictionaries that map each key to its corresponding description.
+
+        2. **Construct the Prompt:**
+            Based on the extracted descriptions, the method builds a detailed prompt that clearly lists all input
+            fields (with their descriptions) and output fields. This prompt is designed to guide the language model,
+            ensuring that it understands what inputs are provided and what outputs are expected.
+
+        3. **Configure the Language Model:**
+            The method initializes and configures a language model using the dspy library. The agent's model
+            (e.g., "openai/gpt-4o") is used to set up the language model, ensuring that it is ready to process the prompt.
+
+        4. **Execute the Task:**
+            Depending on whether the agent has been configured with additional tools:
+                - **With Tools:** A ReAct task is instantiated. This task interleaves reasoning and tool usage,
+                allowing the agent to leverage external functionalities during execution.
+                - **Without Tools:** A Predict task is used for a straightforward generation based on the prompt.
+
+        5. **Process the Result:**
+            After execution, the method attempts to convert the result to a dictionary. It also ensures that each
+            expected input key is present in the output (by setting a default value from the inputs if necessary).
+
+        6. **Error Handling:**
+            Any exceptions raised during the process are caught, logged (or printed), and then re-raised to allow
+            higher-level error handling. This ensures that errors are not silently ignored and can be properly diagnosed.
+
+        **Arguments:**
+            inputs (dict[str, Any]): A dictionary containing all the resolved input values required by the agent.
+                These inputs are typically obtained from the global context or from the output of a previous agent.
+
+        **Returns:**
+            dict[str, Any]: A dictionary containing the output generated by the agent. This output adheres to the
+                agent's declared output fields and includes any fallback values for missing inputs.
+
+        **Usage Example:**
+
+            Suppose an agent is declared with the following configuration:
+                input = "query: str | The search query, context: dict | Additional context"
+                output = "idea: str | The generated software project idea"
+
+            When invoked with:
+                inputs = {"query": "build an app", "context": {"previous_idea": "messaging app"}}
+
+            The method will:
+                - Parse the descriptions to create:
+                    input_descriptions = {"query": "The search query", "context": "Additional context"}
+                    output_descriptions = {"idea": "The generated software project idea"}
+                - Construct a prompt that lists these inputs and outputs clearly.
+                - Configure the language model and execute the appropriate task (ReAct if tools are provided, otherwise Predict).
+                - Return a dictionary similar to:
+                    {"idea": "A fun app idea based on ...", "query": "build an app", "context": {"previous_idea": "messaging app"}}
+        """
+        try:
+            self.__dspy_signature = self.create_dspy_signature_class(
+                self.name, self.description, f"{self.input} -> {self.output}"
+            )
+            # Initialize the language model.
+            self._configure_language_model()
+            # Select the appropriate DSPy task based on tool availability.
+            agent_task = self._select_task(self.__dspy_signature)
+            # Execute the task with the provided inputs.
+            result = agent_task(**inputs)
+            # Process the result and ensure fallback values for missing keys.
+            result = self._process_result(result, inputs)
+            return result
+        except Exception as eval_error:
+            logger.error(
+                f"Error during evaluation in agent '{self.name}': {eval_error}"
+            )
+            raise
+
+    async def run(self, inputs: dict[str, Any]) -> dict[str, Any]:
+        """Run the agent with the given inputs and return its generated output.
+
+        This method represents the primary execution flow for a FlockAgent and performs the following
+        lifecycle steps in sequence:
+
+        1. **Initialization:**
+            Calls the `initialize(inputs)` hook to perform any necessary pre-run setup, such as input
+            validation, resource allocation, or logging. This ensures that the agent is properly configured
+            before processing begins.
+
+        2. **Evaluation:**
+            Invokes the internal `_evaluate(inputs)` method, which constructs a detailed prompt (incorporating
+            input and output descriptions), configures the underlying language model via dspy, and executes the
+            main task (using a ReAct task if tools are provided, or a Predict task otherwise).
+
+        3. **Termination:**
+            Calls the `terminate(inputs, result)` hook after evaluation, allowing the agent to clean up any
+            resources or perform post-run actions (such as logging the output).
+
+        4. **Output Return:**
+            Returns a dictionary containing the agent's output. This output conforms to the agent's declared
+            output fields and may include any default or fallback values for missing inputs.
+
+        If an error occurs during any of these steps, the `on_error(error, inputs)` hook is invoked to handle
+        the exception (for instance, by logging detailed error information). The error is then re-raised, ensuring
+        that higher-level error management can address the failure.
+
+        **Arguments:**
+            inputs (dict[str, Any]): A dictionary containing the resolved input values required by the agent.
+                These inputs are typically derived from the agent's declared input signature and may include data
+                provided by previous agents or from a global context.
+
+        **Returns:**
+            dict[str, Any]: A dictionary containing the output generated by the agent. The output structure
+            adheres to the agent's declared output fields.
+
+        **Example:**
+            Suppose an agent is defined with:
+                input  = "query: str | The search query, context: dict | Additional context"
+                output = "result: str | The generated idea"
+            When executed with:
+                inputs = {"query": "build a chatbot", "context": {"user": "Alice"}}
+            The method might return:
+                {"result": "A conversational chatbot that uses AI to...", "query": "build a chatbot", "context": {"user": "Alice"}}
+        """
+        try:
+            await self.initialize(inputs)
+            result = await self.evaluate(inputs)
+            await self.terminate(inputs, result)
+            return result
+        except Exception as run_error:
+            await self.on_error(run_error, inputs)
+            logger.error(f"Error running agent '{self.name}': {run_error}")
+            raise
+
+    async def run_temporal(self, inputs: dict[str, Any]) -> dict[str, Any]:
+        """Execute this agent via a Temporal workflow for enhanced fault tolerance and asynchronous processing.
+
+        This method enables remote execution of the agent within a Temporal environment, leveraging Temporal's
+        capabilities for persistence, retries, and distributed error handling. The workflow encapsulates the agent's
+        logic so that it can run on Temporal workers, providing robustness and scalability in production systems.
+
+        The method performs these steps:
+        1. **Connect to Temporal:**
+            Establishes a connection to the Temporal server using a Temporal client configured with the appropriate
+            namespace (default is "default").
+        2. **Serialization:**
+            Serializes the agent instance (via `to_dict()`) along with the provided inputs. This step converts any
+            callable objects (e.g., lifecycle hooks or tools) into a storable format using cloudpickle.
+        3. **Activity Invocation:**
+            Triggers a designated Temporal activity (e.g., `run_flock_agent_activity`) by passing the serialized agent
+            data and inputs. The Temporal activity is responsible for executing the agent's logic in a fault-tolerant
+            manner.
+        4. **Return Output:**
+            Awaits and returns the resulting output from the Temporal workflow as a dictionary, consistent with the
+            output structure of the local `run()` method.
+
+        If any error occurs during these steps, the error is logged and re-raised to ensure that failure is properly
+        handled by higher-level error management systems.
+
+        **Arguments:**
+            inputs (dict[str, Any]): A dictionary containing the resolved inputs required by the agent, similar to those
+                provided to the local `run()` method.
+
+        **Returns:**
+            dict[str, Any]: A dictionary containing the output produced by the agent after remote execution via Temporal.
+                The output format is consistent with that of the local `run()` method.
+
+        **Example:**
+            Given an agent defined with:
+                input  = "query: str | The search query, context: dict | Additional context"
+                output = "result: str | The generated idea"
+            Calling:
+                result = await agent.run_temporal({"query": "analyze data", "context": {"source": "sales"}})
+            will execute the agent on a Temporal worker and return the output in a structured dictionary format.
+        """
+        try:
+            from temporalio.client import Client
+
+            from flock.workflow.agent_activities import run_flock_agent_activity
+            from flock.workflow.temporal_setup import run_activity
+
+            client = await Client.connect("localhost:7233", namespace="default")
+            agent_data = self.to_dict()
+            inputs_data = inputs
+
+            result = await run_activity(
+                client,
+                self.name,
+                run_flock_agent_activity,
+                {"agent_data": agent_data, "inputs": inputs_data},
+            )
+            return result
+
+        except Exception as temporal_error:
+            logger.error(
+                f"Error running Temporal workflow for agent '{self.name}': {temporal_error}"
+            )
+            raise
+
+    def to_dict(self) -> dict[str, Any]:
+        """Serialize the FlockAgent instance to a dictionary.
+
+        This method converts the entire agent instance—including its configuration, state, and lifecycle hooks—
+        into a dictionary format. It uses cloudpickle to serialize any callable objects (such as functions or
+        methods), converting them into hexadecimal string representations. This ensures that the agent can be
+        easily persisted, transmitted, or logged as JSON.
+
+        The serialization process is recursive:
+        - If a field is a callable (and not a class), it is serialized using cloudpickle.
+        - Lists and dictionaries are processed recursively to ensure that all nested callables are properly handled.
+
+        **Returns:**
+            dict[str, Any]: A dictionary representing the FlockAgent, which includes all of its configuration data.
+            This dictionary is suitable for storage, debugging, or transmission over the network.
+
+        **Example:**
+            For an agent defined as:
+                name = "idea_agent",
+                model = "openai/gpt-4o",
+                input = "query: str | The search query, context: dict | The full conversation context",
+                output = "idea: str | The generated idea"
+            Calling `agent.to_dict()` might produce:
+                {
+                    "name": "idea_agent",
+                    "model": "openai/gpt-4o",
+                    "input": "query: str | The search query, context: dict | The full conversation context",
+                    "output": "idea: str | The generated idea",
+                    "tools": ["<serialized tool representation>"],
+                    "use_cache": False,
+                    "hand_off": None,
+                    "termination": None,
+                    ...
+                }
+        """
+
+        def convert_callable(obj: Any) -> Any:
+            if callable(obj) and not isinstance(obj, type):
+                return cloudpickle.dumps(obj).hex()
+            if isinstance(obj, list):
+                return [convert_callable(item) for item in obj]
+            if isinstance(obj, dict):
+                return {k: convert_callable(v) for k, v in obj.items()}
+            return obj
+
+        data = self.dict()
+        return convert_callable(data)
+
+    @classmethod
+    def from_dict(cls, data: dict[str, Any]) -> "FlockAgent":
+        """Deserialize a FlockAgent instance from a dictionary.
+
+        This class method reconstructs a FlockAgent from its serialized dictionary representation, as produced
+        by the `to_dict()` method. It recursively processes the dictionary to convert any serialized callables
+        (stored as hexadecimal strings via cloudpickle) back into executable callable objects.
+
+        **Arguments:**
+            data (dict[str, Any]): A dictionary representation of a FlockAgent, typically produced by `to_dict()`.
+                The dictionary should contain all configuration fields and state information necessary to fully
+                reconstruct the agent.
+
+        **Returns:**
+            FlockAgent: An instance of FlockAgent reconstructed from the provided dictionary. The deserialized agent
+            will have the same configuration, state, and behavior as the original instance.
+
+        **Example:**
+            Suppose you have the following dictionary:
+                {
+                    "name": "idea_agent",
+                    "model": "openai/gpt-4o",
+                    "input": "query: str | The search query, context: dict | The full conversation context",
+                    "output": "idea: str | The generated idea",
+                    "tools": ["<serialized tool representation>"],
+                    "use_cache": False,
+                    "hand_off": None,
+                    "termination": None,
+                    ...
+                }
+            Then, calling:
+                agent = FlockAgent.from_dict(data)
+            will return a FlockAgent instance with the same properties and behavior as when it was originally serialized.
+        """
+
+        def convert_callable(obj: Any) -> Any:
+            if isinstance(obj, str) and len(obj) > 2:
+                try:
+                    return cloudpickle.loads(bytes.fromhex(obj))
+                except Exception:
+                    return obj
+            if isinstance(obj, list):
+                return [convert_callable(item) for item in obj]
+            if isinstance(obj, dict):
+                return {k: convert_callable(v) for k, v in obj.items()}
+            return obj
+
+        converted = convert_callable(data)
+        return cls(**converted)

flock/core/logging/__init__.py

@@ -1,18 +1,2 @@
-"""Flock logging system with Rich integration and Temporal compatibility."""
-
-from flock.core.logging.error_handler import error_handler
-from flock.core.logging.formatters import PerformanceFormatter, StructuredFormatter
-from flock.core.logging.handlers import live_update_handler, performance_handler
-from flock.core.logging.logger import flock_logger
-
-# Install the Rich error handler by default
-error_handler.install()
-
-__all__ = [
-    "PerformanceFormatter",
-    "StructuredFormatter",
-    "error_handler",
-    "flock_logger",
-    "live_update_handler",
-    "performance_handler",
-]
+"""Flock logging system with Rich integration and structured logging support."""
+

flock/core/logging/formatters/base_formatter.py

@@ -0,0 +1,36 @@
+from abc import ABC, abstractmethod
+from dataclasses import dataclass, field
+from typing import Any
+
+
+class BaseFormatter(ABC):
+    def __init__(self, max_length: int = 1000) -> None:
+        self.max_length = max_length
+
+    def display(
+        self,
+        result: dict[str, Any],
+        agent_name: str,
+        wait: bool = False,
+    ) -> None:
+        self.display_result(result, agent_name)
+        if wait:
+            input("Press Enter to continue...")
+
+    @abstractmethod
+    def display_result(self, result: dict[str, Any], agent_name: str) -> None:
+        """Display an agent's result."""
+        raise NotImplementedError
+
+    @abstractmethod
+    def display_data(self, data: dict[str, Any]) -> None:
+        """Display arbitrary data."""
+        raise NotImplementedError
+
+
+@dataclass
+class FormatterOptions:
+    formatter: type[BaseFormatter] = field(default=None)
+    wait_for_input: bool = field(default=False)
+    max_length: int = field(default=1000)
+    settings: Any | None = field(default=None)

flock/core/logging/formatters/formatter_factory.py

@@ -0,0 +1,38 @@
+from dataclasses import dataclass
+from typing import Any
+
+from flock.core.logging.formatters.base_formatter import (
+    BaseFormatter,
+    FormatterOptions,
+)
+from flock.core.logging.formatters.pprint_formatter import PrettyPrintFormatter
+from flock.core.logging.formatters.rich_formatters import RichTables
+from flock.core.logging.formatters.themed_formatter import (
+    ThemedAgentResultFormatter,
+)
+
+
+@dataclass
+class FormatterOptions:
+    formatter: type[BaseFormatter]
+    wait_for_input: bool = False
+    settings: dict[str, Any] | None = None
+
+
+class FormatterFactory:
+    _formatter_map = {
+        PrettyPrintFormatter: PrettyPrintFormatter,
+        RichTables: RichTables,
+        ThemedAgentResultFormatter: ThemedAgentResultFormatter,
+    }
+
+    @staticmethod
+    def create_formatter(options: FormatterOptions) -> BaseFormatter:
+        formatter_cls = options.formatter
+        max_length = options.max_length
+        if formatter_cls in FormatterFactory._formatter_map:
+            formatter = FormatterFactory._formatter_map[formatter_cls]
+            if options.settings:
+                return formatter(max_length=max_length, **options.settings)
+            return formatter(max_length=max_length)
+        raise ValueError(f"Unknown formatter: {formatter_cls}")

flock/core/logging/formatters/pprint_formatter.py

@@ -0,0 +1,18 @@
+from typing import Any
+
+from flock.core.logging.formatters.base_formatter import BaseFormatter
+
+
+class PrettyPrintFormatter(BaseFormatter):
+    def display_result(self, result: dict[str, Any], agent_name: str, **kwargs) -> None:
+        """Print an agent's result using Rich formatting."""
+        from devtools import pprint
+
+        pprint(agent_name)
+        pprint(result)
+
+    def display_data(self, data: dict[str, Any], **kwargs) -> None:
+        """Print an agent's result using Rich formatting."""
+        from devtools import pprint
+
+        pprint(data)