aip-agents-binary 0.5.20__py3-none-any.whl → 0.5.22__py3-none-any.whl

aip_agents/utils/pii/pii_handler.pyi

@@ -0,0 +1,96 @@
+from _typeshed import Incomplete
+from aip_agents.utils.logger import LoggerManager as LoggerManager
+from aip_agents.utils.pii.uuid_deanonymizer_mapping import UUIDDeanonymizerMapping as UUIDDeanonymizerMapping
+from enum import Enum
+from gllm_privacy.pii_detector import TextAnalyzer, TextAnonymizer
+from typing import Any
+
+TextAnalyzer = Any
+TextAnonymizer = Any
+GDPLabsNerApiRemoteRecognizer = Any
+
+class _Operation(str, Enum):
+    ANONYMIZE: str
+    DEANONYMIZE: str
+
+logger: Incomplete
+NER_API_URL_ENV_VAR: str
+NER_API_KEY_ENV_VAR: str
+NER_API_TIMEOUT: int
+EXCLUDED_ENTITIES: Incomplete
+DEFAULT_SUPPORTED_ENTITIES: Incomplete
+
+class ToolPIIHandler:
+    """Handles PII masking/demasking for tool calling.
+
+    Tag replacement based on runner-provided mappings always works. Optional
+    NER-powered masking/de-masking is only enabled when NER_API_URL and
+    NER_API_KEY environment variables are set.
+
+    Attributes:
+        flat_pii_mapping: Flat mapping from runner service (tag → value)
+        text_analyzer: GLLM Privacy TextAnalyzer instance
+        text_anonymizer: GLLM Privacy TextAnonymizer instance
+        enable_ner: Whether NER is enabled
+    """
+    flat_pii_mapping: dict[str, str]
+    enable_ner: bool
+    text_analyzer: TextAnalyzer | None
+    text_anonymizer: TextAnonymizer | None
+    def __init__(self, pii_mapping: dict[str, str] | None = None, ner_api_url: str | None = None, ner_api_key: str | None = None) -> None:
+        """Initialize PII handler (private - use create_if_enabled() instead).
+
+        Initializes GLLM Privacy components (TextAnalyzer, TextAnonymizer) if NER credentials
+        are provided. Creates dual recognizers for Indonesian and English languages.
+        Pre-loads any existing PII mappings into the anonymizer's internal state.
+
+        Args:
+            pii_mapping: Existing PII mapping from runner service (flat format: tag -> value)
+            ner_api_url: NER API endpoint URL
+            ner_api_key: NER API authentication key
+        """
+    @classmethod
+    def create_if_enabled(cls, pii_mapping: dict[str, str] | None = None) -> ToolPIIHandler | None:
+        """Create ToolPIIHandler when mappings or NER configuration exist.
+
+        Args:
+            pii_mapping: Existing PII mapping from runner service
+
+        Returns:
+            ToolPIIHandler instance when mapping or NER config is available, None otherwise
+        """
+    @classmethod
+    def create_mapping_only(cls, pii_mapping: dict[str, str] | None = None) -> ToolPIIHandler | None:
+        """Create ToolPIIHandler in mapping-only mode (no NER).
+
+        Args:
+            pii_mapping: Existing PII mapping from runner service
+
+        Returns:
+            ToolPIIHandler instance when mapping exists, None otherwise
+        """
+    def deanonymize_tool_args(self, args: dict[str, Any]) -> dict[str, Any]:
+        """Replace PII tags in tool arguments with real values.
+
+        Recursively processes dictionaries, lists, and strings to replace all PII tags
+        (e.g., '<EMAIL_1>') with their corresponding real values from flat_pii_mapping.
+
+        Args:
+            args: Tool arguments that may contain PII tags
+
+        Returns:
+            Arguments with tags replaced by real values
+        """
+    def anonymize_tool_output(self, output: Any) -> tuple[Any, dict[str, str]]:
+        """Mask PII values in tool output.
+
+        Handles string and dictionary outputs. For strings, uses two-phase anonymization:
+        first masks known PII, then detects new PII via NER. For dictionaries, recursively
+        processes all string values. Returns updated mapping with any newly discovered PII.
+
+        Args:
+            output: Tool output that may contain PII values (string, dict, or other)
+
+        Returns:
+            Tuple of (anonymized_output, updated_flat_pii_mapping)
+        """

aip_agents/utils/pii/pii_helper.pyi

@@ -0,0 +1,78 @@
+from _typeshed import Incomplete
+from aip_agents.utils.logger import LoggerManager as LoggerManager
+from typing import Any
+
+logger: Incomplete
+
+def normalize_enable_pii(enable_pii: Any) -> bool | None:
+    """Normalize enable_pii value from agent configuration.
+
+    Args:
+        enable_pii: Raw enable_pii value from agent configuration.
+
+    Returns:
+        The normalized enable_pii flag when explicitly set (True/False), otherwise None.
+    """
+def add_pii_mappings(left: dict[str, str] | None, right: dict[str, str] | None) -> dict[str, str]:
+    """Reducer function to merge PII mappings from multiple sources.
+
+    This is a LangGraph reducer function that merges PII mappings from:
+    - Parent agent's initial mapping
+    - Tool outputs with newly discovered PII
+    - Subagent responses with their discovered PII
+
+    Args:
+        left: Existing PII mapping (or None)
+        right: New PII mapping to merge (or None)
+
+    Returns:
+        Merged PII mapping dictionary
+
+    Note:
+        - Right (new) mappings take precedence over left (existing)
+        - Handles None/non-dict cases gracefully
+        - Preserves all unique PII tags
+        - Returns empty dict if both inputs are None/empty
+    """
+def extract_pii_mapping_from_agent_response(result: Any) -> dict[str, str] | None:
+    """Extract PII mapping from subagent response.
+
+    Used by DelegationToolManager to propagate PII mappings from subagents
+    back to parent agents.
+
+    Args:
+        result: The result returned by the delegated agent
+
+    Returns:
+        PII mapping dictionary if found, None otherwise
+
+    Note:
+        - Checks if result is a dict
+        - Extracts 'full_final_state' from result
+        - Extracts 'pii_mapping' from full_final_state
+        - Validates mapping is a non-empty dict
+        - Returns None if any step fails
+    """
+def deanonymize_final_response_content(content: str, is_final_response: bool, metadata: dict[str, Any] | None) -> str:
+    """Deanonymize final response content using PII mapping from metadata.
+
+    Args:
+        content: Final response content that may contain PII tags.
+        is_final_response: Flag indicating whether this message is a final response.
+        metadata: Optional metadata dict (or event payload containing ``metadata``) with
+            ``pii_mapping`` tag-to-value mapping.
+
+    Returns:
+        Content string with PII tags replaced by real values when applicable.
+    """
+def anonymize_final_response_content(content: str, metadata: dict[str, Any] | None) -> str:
+    """Anonymize final response content using PII mapping from metadata.
+
+    Args:
+        content: Final response content that may contain real PII values.
+        metadata: Metadata dict (or event payload containing ``metadata``) with
+            ``pii_mapping`` tag-to-value mapping.
+
+    Returns:
+        Content string with real PII values replaced by their PII tags when mapping is present.
+    """

aip_agents/utils/pii/uuid_deanonymizer_mapping.pyi

@@ -0,0 +1,73 @@
+from _typeshed import Incomplete
+from aip_agents.utils.constants import DEFAULT_PII_TAG_NAMESPACE as DEFAULT_PII_TAG_NAMESPACE
+from gllm_privacy.pii_detector.utils.deanonymizer_mapping import DeanonymizerMapping, MappingDataType
+from typing import Any
+
+MappingDataType = dict[str, dict[str, str]]
+
+class DeanonymizerMapping:
+    """Fallback deanonymizer mapping when optional dependency is missing.
+
+        This class exists only to keep the module importable when `gllm-privacy`
+        is not installed.
+        """
+    def __init__(self, *args: Any, **kwargs: Any) -> None:
+        """Initialize the mapping.
+
+            Raises:
+                ImportError: Always raised because `gllm-privacy` is required.
+            """
+
+logger: Incomplete
+
+def format_operator_with_uuid(operator_name: str, uuid_suffix: str) -> str:
+    """Format the operator name with a UUID suffix.
+
+    Args:
+        operator_name: The operator name.
+        uuid_suffix: The UUID suffix to append.
+
+    Returns:
+        The formatted operator name with UUID suffix.
+    """
+
+class UUIDDeanonymizerMapping(DeanonymizerMapping):
+    """Class to store the deanonymizer mapping with UUID suffixes.
+
+    This class extends DeanonymizerMapping to use UUID suffixes instead of
+    sequential numbers for differentiating multiple entities of the same type.
+
+    Attributes:
+        mapping: The deanonymizer mapping.
+        skip_format_duplicates: Whether to skip formatting duplicated operators.
+        uuid_length: The length of the UUID suffix to use (default: 8).
+    """
+    mapping: Incomplete
+    skip_format_duplicates: Incomplete
+    uuid_length: Incomplete
+    def __init__(self, mapping: MappingDataType | None = None, skip_format_duplicates: bool = False, uuid_length: int = 8) -> None:
+        """Initialize UUIDDeanonymizerMapping.
+
+        Args:
+            mapping: The deanonymizer mapping. If None, creates an empty defaultdict(dict).
+            skip_format_duplicates: Whether to skip formatting duplicated operators.
+            uuid_length: The length of the UUID suffix to use (default: 8).
+        """
+    def update(self, new_mapping: MappingDataType, use_uuid_suffix: bool | None = None) -> None:
+        """Update the deanonymizer mapping with new values using UUID suffixes.
+
+        Duplicated values will not be added. If there are multiple entities of the same type,
+        the mapping will include a UUID suffix to differentiate them. For example, if there are
+        two names in the input text, the mapping will include NAME_<uuid1> and NAME_<uuid2>.
+
+        Args:
+            new_mapping: The new mapping to be added to the existing deanonymizer mapping.
+            use_uuid_suffix: Whether to apply UUID suffixes to keys.
+                If True, keys will always be formatted with UUID suffixes.
+                If False, keys will be used as-is without UUID formatting.
+                If None, behavior falls back to the instance configuration via
+                skip_format_duplicates (preserving existing behavior).
+
+        Returns:
+            None
+        """

aip_agents/utils/reference_helper.pyi

@@ -0,0 +1,81 @@
+from _typeshed import Incomplete
+from aip_agents.utils.logger import get_logger as get_logger
+from aip_agents.utils.metadata_helper import MetadataFieldKeys as MetadataFieldKeys
+from gllm_core.schema import Chunk
+from langgraph.types import Command
+from typing import Any
+
+logger: Incomplete
+SAVE_OUTPUT_HISTORY_ATTR: str
+FORMAT_AGENT_REFERENCE: str
+
+def extract_references_from_tool(tool: Any, tool_output: Any) -> list[Chunk]:
+    """Extract reference data from tools that support it.
+
+    Extraction priority:
+    1. Direct tool references via _format_agent_reference (preferred)
+    2. Command.update references (fallback for delegation tools)
+
+    Args:
+        tool: The tool instance to extract references from
+        tool_output: The output from the tool execution
+    Returns:
+        List of deduplicated Chunk objects containing reference data
+
+    Note:
+        - Never raises exceptions; logs warnings for issues
+        - Direct tool references take precedence over Command references
+    """
+def extract_references_from_command_update(command: Command) -> list[Any]:
+    """Extract references from a Command object's update dictionary.
+
+    Args:
+        command: A Command object potentially containing references in its update dict.
+
+    Returns:
+        List of reference data (not yet validated/deduplicated)
+
+    Note:
+        - Never raises exceptions; logs warnings for malformed data
+        - Skips non-Chunk items with warning log
+    """
+def validate_references(references: list[Any]) -> list[Chunk]:
+    """Validate and deduplicate reference data.
+
+    Args:
+        references: List of reference data (expected to be Chunk objects).
+
+    Returns:
+        List of validated, deduplicated Chunk objects by content.
+    """
+def serialize_references_for_metadata(references: list[Any]) -> list[dict[str, Any]]:
+    """Serialize references for inclusion in A2A metadata.
+
+    Args:
+        references: List of reference objects (typically Chunk objects).
+
+    Returns:
+        List of serialized reference dictionaries.
+    """
+def add_references_chunks(left: list[Chunk], right: list[Chunk]) -> list[Chunk]:
+    """Reducer function to accumulate reference data from multiple tool calls.
+
+    This is a LangGraph reducer function that should be forgiving and handle
+    edge cases gracefully. Non-Chunk items are filtered out.
+
+    Args:
+        left: Existing list of reference chunks (or None/non-list)
+        right: New list of reference chunks to add (or None/non-list)
+
+    Returns:
+        Combined list of valid Chunk objects
+    """
+def extract_references_from_agent_response(result: Any) -> list[dict[str, Any]] | None:
+    """Extract references from agent response for delegation tools.
+
+    Args:
+        result: The result returned by the delegated agent.
+
+    Returns:
+        List of reference chunks if found, None otherwise.
+    """

aip_agents/utils/sse_chunk_transformer.pyi

@@ -0,0 +1,166 @@
+from _typeshed import Incomplete
+from aip_agents.schema.a2a import A2AEvent
+from collections.abc import AsyncGenerator
+from enum import StrEnum
+from typing import Any
+
+__all__ = ['SSEChunkTransformer', 'TaskState', 'ChunkStatus', 'ChunkReason', 'ChunkFieldKeys']
+
+class TaskState(StrEnum):
+    """Task state values for SSE chunks."""
+    WORKING: str
+    COMPLETED: str
+    FAILED: str
+
+class ChunkStatus(StrEnum):
+    """Status values for SSE chunks."""
+    SUCCESS: str
+    ERROR: str
+
+class ChunkReason(StrEnum):
+    """Reason codes for special chunk states."""
+    EMPTY_PAYLOAD: str
+
+class ChunkFieldKeys(StrEnum):
+    """Field name constants for SSE chunk structure."""
+    STATUS: str
+    TASK_STATE: str
+    CONTENT: str
+    EVENT_TYPE: str
+    FINAL: str
+    METADATA: str
+    TIMESTAMP: str
+    TASK_ID: str
+    CONTEXT_ID: str
+    ARTIFACTS: str
+    REASON: str
+
+class SSEChunkTransformer:
+    '''Transforms A2AEvent stream to SSE-compatible output.
+
+    This class converts events from arun_a2a_stream into the normalized dict format
+    matching A2AConnector.astream_to_agent output.
+
+    Lifecycle:
+        Single-stream instance. Must NOT be reused across concurrent streams.
+        Each arun_sse_stream call creates a fresh instance.
+
+    Attributes:
+        task_id: Optional task identifier for the stream.
+        context_id: Optional context identifier for the stream.
+
+    Example:
+        >>> transformer = SSEChunkTransformer(task_id="task-123")
+        >>> async for chunk in transformer.transform_stream(agent.arun_a2a_stream("query")):
+        ...     print(chunk)
+    '''
+    task_id: Incomplete
+    context_id: Incomplete
+    def __init__(self, task_id: str | None = None, context_id: str | None = None, pii_mapping: dict[str, str] | None = None) -> None:
+        """Initialize the transformer with optional task and context IDs.
+
+        Args:
+            task_id: Optional task identifier for the stream.
+            context_id: Optional context identifier for the stream.
+            pii_mapping: Optional PII mapping to inject into each chunk's metadata.
+        """
+    @staticmethod
+    def normalize_metadata_enums(data: Any) -> Any:
+        """Recursively convert enum keys/values to their string values.
+
+        This is a pure normalization utility that converts any enum instances
+        (MetadataFieldKeys, Kind, Status, etc.) to their .value strings.
+
+        Args:
+            data: Dict, list, or value that may contain enum keys/values.
+
+        Returns:
+            Normalized data with all enums converted to their .value strings.
+        """
+    @staticmethod
+    def normalize_event_type_value(event_type: Any) -> str | None:
+        """Convert A2AStreamEventType enum to string.
+
+        Args:
+            event_type: Event type (enum, string, or None).
+
+        Returns:
+            String value of the event type, or None if invalid.
+        """
+    @staticmethod
+    def create_artifact_hash(artifact: dict[str, Any]) -> str:
+        """Create a stable hash for artifact deduplication.
+
+        Uses name, content_type, mime_type, and file_data for hashing,
+        excluding artifact_id which may be randomly generated.
+
+        Args:
+            artifact: Artifact dict with name, content_type, mime_type, and optionally file_data.
+
+        Returns:
+            SHA256 hexdigest hash string for deduplication.
+        """
+    @staticmethod
+    def extract_tool_outputs(tool_calls: list[dict[str, Any]]) -> list[str]:
+        """Extract human-readable output strings from tool calls.
+
+        Args:
+            tool_calls: List of tool call dictionaries.
+
+        Returns:
+            List of human-readable output strings.
+        """
+    @staticmethod
+    def format_tool_output(output: Any, tool_name: str) -> str:
+        """Format a single tool output for display.
+
+        Args:
+            output: The tool output to format.
+            tool_name: The name of the tool.
+
+        Returns:
+            The formatted output string.
+        """
+    @staticmethod
+    def apply_hitl_content_override(content: str | None, event_type_str: str, metadata: dict[str, Any]) -> str | None:
+        """Apply HITL content override when HITL is active and tool results are available.
+
+        This method overrides the content with human-readable tool output when HITL
+        is active, matching A2AConnector behavior.
+
+        Args:
+            content: The original content/status message.
+            event_type_str: The type of event being processed (normalized string).
+            metadata: The metadata dictionary containing tool_info and hitl flag.
+
+        Returns:
+            The original content or human-readable tool output if HITL is active.
+        """
+    def transform_event(self, event: A2AEvent) -> dict[str, Any]:
+        """Transform a single A2AEvent to SSE chunk format.
+
+        Converts the A2AEvent structure to the normalized SSE chunk format,
+        relocating fields like tool_info and thinking_and_activity_info into
+        metadata, and normalizing enum values to strings.
+
+        Args:
+            event: Single A2AEvent dict from arun_a2a_stream.
+
+        Returns:
+            SSEChunk dict with normalized structure.
+        """
+    async def transform_stream(self, stream: AsyncGenerator[A2AEvent, None]) -> AsyncGenerator[dict[str, Any], None]:
+        """Transform A2AEvent stream to SSE-compatible chunks.
+
+        Wraps the input stream and transforms each event, handling artifact
+        deduplication and time tracking across the stream.
+
+        Args:
+            stream: Async generator yielding A2AEvent dicts.
+
+        Yields:
+            SSEChunk dicts with normalized structure.
+
+        Raises:
+            Exceptions from underlying stream propagate to caller.
+        """

aip_agents/utils/step_limit_manager.pyi

@@ -0,0 +1,112 @@
+from _typeshed import Incomplete
+from aip_agents.schema.step_limit import MaxDelegationDepthExceededError as MaxDelegationDepthExceededError, MaxStepsExceededError as MaxStepsExceededError, StepLimitConfig as StepLimitConfig, StepLimitErrorResponse as StepLimitErrorResponse, StepLimitErrorType as StepLimitErrorType
+from dataclasses import dataclass, field
+from typing import Any
+
+@dataclass
+class StepExecutionContext:
+    """Runtime context for tracking step execution and delegation depth.
+
+    Attributes:
+        current_step: Current step number (0-indexed).
+        delegation_depth: Current depth in delegation chain (0 for root agent).
+        remaining_step_budget: Steps remaining before limit is hit.
+        delegation_chain: List of agent names in the delegation chain.
+    """
+    current_step: int = ...
+    delegation_depth: int = ...
+    remaining_step_budget: int | None = ...
+    delegation_chain: list[str] = field(default_factory=list)
+
+class StepLimitManager:
+    """Manages step and delegation limit enforcement during agent execution.
+
+    This manager integrates with LangGraph's existing step mechanisms and adds
+    delegation depth tracking and budget propagation.
+
+    Attributes:
+        config: Step limit configuration.
+        context: Current execution context.
+    """
+    config: Incomplete
+    context: Incomplete
+    def __init__(self, config: StepLimitConfig | None = None, initial_delegation_depth: int = 0, parent_step_budget: int | None = None) -> None:
+        """Initialize step limit manager.
+
+        Args:
+            config: Optional step limit configuration. Uses defaults if None.
+            initial_delegation_depth: Starting delegation depth (from parent).
+            parent_step_budget: Remaining step budget inherited from parent agent.
+        """
+    def check_step_limit(self, agent_name: str = 'agent', count: int = 1) -> None:
+        """Check if taking 'count' steps would exceed limit.
+
+        Args:
+            agent_name: Name of the agent to identify in error message.
+            count: Number of steps to check (useful for parallel tool batches).
+
+        Raises:
+            MaxStepsExceededError: If max_steps limit is exceeded.
+        """
+    def check_delegation_depth(self, target_agent_name: str) -> None:
+        """Check if delegation to target agent would exceed depth limit.
+
+        Args:
+            target_agent_name: Name of the agent to delegate to.
+
+        Raises:
+            MaxDelegationDepthExceededError: If delegation depth limit exceeded.
+        """
+    def increment_step(self, count: int = 1) -> None:
+        """Increment step counter and update remaining budget.
+
+        Args:
+            count: Number of steps to consume (defaults to 1).
+        """
+    def get_child_budget(self, child_max_steps: int | None = None) -> int:
+        """Calculate step budget to allocate to child agent.
+
+        Algorithm:
+        1. If remaining_step_budget is None (root with no limit), use config.max_steps - 1
+        2. If remaining_step_budget <= 1, return 0 (no budget left for child)
+        3. Calculate child_budget = remaining_step_budget - 1 (reserve 1 for parent)
+        4. If child has own max_steps config, return min(child_budget, child.max_steps)
+        5. Otherwise return child_budget
+
+        Args:
+            child_max_steps: Optional child agent's own max_steps limit.
+
+        Returns:
+            Step budget for child agent, accounting for parent's continuation.
+            Returns 0 if no budget available for child.
+
+        Edge Cases:
+            - remaining=1: Returns 0 (parent needs the last step)
+            - remaining=None: Uses config.max_steps - 1
+            - child has own limit: Returns min(calculated_budget, child_limit)
+        """
+    def add_to_delegation_chain(self, agent_name: str) -> None:
+        """Add agent to delegation chain for tracking.
+
+        Args:
+            agent_name: Name of the agent being delegated to.
+        """
+    @classmethod
+    def from_state(cls, state: dict[str, Any], config: StepLimitConfig | None = None) -> StepLimitManager:
+        """Create manager from LangGraph state.
+
+        Args:
+            state: LangGraph agent state containing remaining_steps, etc.
+            config: Optional step limit configuration.
+
+        Returns:
+            Initialized step limit manager.
+        """
+    def set_context(self) -> None:
+        """Set context variables for downstream consumption (e.g. by delegation tools)."""
+    def to_state_update(self) -> dict[str, Any]:
+        """Convert current context to LangGraph state update.
+
+        Returns:
+            Dictionary of state fields to update.
+        """

aip_agents/utils/token_usage_helper.pyi

@@ -0,0 +1,60 @@
+from langchain_core.messages.ai import AIMessage, UsageMetadata
+from langchain_core.messages.tool import ToolMessage as ToolMessage
+from langgraph.types import Command
+from typing import Any
+
+USAGE_METADATA_KEY: str
+TOTAL_USAGE_KEY: str
+STEP_USAGE_KEY: str
+
+def add_usage_metadata(cur_accumulated_token_usage: UsageMetadata | None, new_token_usage: UsageMetadata | None) -> UsageMetadata | None:
+    """Reducer function to accumulate UsageMetadata across agent runs.
+
+    Args:
+        cur_accumulated_token_usage: The current accumulated token usage metadata.
+        new_token_usage: New token usage metadata to add.
+
+    Returns:
+        Accumulated usage metadata or None if both inputs are None.
+    """
+def extract_and_update_token_usage_from_ai_message(ai_message: AIMessage) -> dict[str, Any]:
+    """Extract token usage from AI message and prepare state update.
+
+    Args:
+        ai_message: The AI message containing usage metadata.
+
+    Returns:
+        Dictionary with accumulated_usage_metadata update if usage metadata is available.
+    """
+def extract_token_usage_from_tool_output(tool_output: Any) -> UsageMetadata | None:
+    """Extract token usage from various tool output formats.
+
+    Supports multiple tool output formats:
+    1. Dictionary with 'usage_metadata' field
+    2. Command with 'usage_metadata' attribute
+    3. Any object with 'usage_metadata' attribute
+
+    Args:
+        tool_output: The output from a tool execution.
+
+    Returns:
+        UsageMetadata if found, None otherwise.
+    """
+def extract_token_usage_from_command(command: Command) -> UsageMetadata | None:
+    """Extract token usage from Command object.
+
+    Args:
+        command: The Command object to extract token usage from.
+
+    Returns:
+        UsageMetadata if found, None otherwise.
+    """
+def extract_token_usage_from_agent_response(agent_response: dict[str, Any]) -> UsageMetadata | None:
+    """Extract accumulated token usage from agent response.
+
+    Args:
+        agent_response: The agent response to extract token usage from.
+
+    Returns:
+        UsageMetadata if found, None otherwise.
+    """

{aip_agents_binary-0.5.20.dist-info → aip_agents_binary-0.5.22.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: aip-agents-binary
-Version: 0.5.20
+Version: 0.5.22
 Summary: A library for managing agents in Gen AI applications.
 Author-email: Raymond Christopher <raymond.christopher@gdplabs.id>
 Requires-Python: <3.13,>=3.11
@@ -159,9 +159,9 @@ Navigate to the library's root directory (e.g., `python/aip-agents` if you clone
 python aip_agents/examples/hello_world_langgraph.py
 ```
 
-**LangGraph with BOSA Connector (OpenAI):**
+**LangGraph with GL Connectors (OpenAI):**
 ```bash
-python aip_agents/examples/hello_world_langgraph_bosa_twitter.py
+python aip_agents/examples/hello_world_langgraph_gl_connector_twitter.py
 ```
 
 **LangGraph Streaming (OpenAI):**