adk_core library

Classes

ActivityEvent: Generic activity or status event.
AdkAbortController: Controller used by callers to cancel an ADK invocation.
AdkAbortSignal: Read-only cancellation signal propagated through an invocation.
AdkMetricRecord: One ADK metric sample.
AdkMetricsRecorder: Sink for ADK metric samples.
AgentNode: Node that wraps an ADK BaseAgent.
AgentTool: Tool adapter that executes another agent as a tool call.
App: App container describing root agent, plugins, and runtime options.
ArtifactVersion: Metadata describing one persisted artifact version.
BaseAgent: Base class for all agent implementations.
BaseAgentState: Generic mutable state holder for agent-local state.
BaseArtifactService: Base contract for artifact persistence services.
BaseLlm: Base interface for model adapters used by the runner.
BaseMemoryService: Base contract for memory ingestion and search implementations.
BaseNode: Base class for node-based workflow units.
BasePlugin: Base class for ADK runtime plugins.
BaseSessionService: Contract for session lifecycle and event persistence operations.
BaseTelemetryService: Contract for telemetry trace/span lifecycle services.
BaseTool: Base contract for one callable tool exposed to the model runtime.
CallCodeEvent: Request to execute code.
CodeResultEvent: Result of code execution.
Content: One conversation turn consisting of a role and Part list.
ContentEvent: User-visible text content.
Context: Mutable context for a single agent invocation.
Edge: Directed workflow edge.
ErrorEvent: Runtime error event.
Event: One runtime event emitted during an invocation.
EventActions: Mutable action payload attached to one event.
EventCompaction: Compaction summary metadata attached to an event.
EventsCompactionConfig: Configuration for event compaction behavior.
FileData: File reference payload attached to a Part.
FinishedEvent: Final response marker.
FinishTaskTool: Tool that validates final task output and signals completion.
Frontmatter: L1 skill metadata parsed from SKILL.md frontmatter.
FunctionCall: Function-call payload emitted by model responses.
FunctionCallingConfig: Function-calling configuration attached to tool settings.
FunctionDeclaration: Function-calling declaration exposed to the model.
FunctionNode: Function-backed workflow node.
FunctionResponse: Function-response payload sent back to the model.
FunctionTool: Wraps a Dart callable as a runtime tool.
Gemini: Google Gemini model adapter with REST, interactions, and live support.
GenerateContentConfig: Generation configuration for text/content model calls.
GetSessionConfig: Options for narrowing session retrieval results.
HttpOptions: HTTP configuration for model API requests.
HttpRetryOptions: Retry policy for HTTP model requests.
InlineData: Inline binary payload attached to a Part.
InMemoryAdkMetricsRecorder: Metrics recorder that stores samples in memory for tests and adapters.
InMemoryArtifactService: Artifact service backed by process-local memory maps.
InMemoryMemoryService: Memory service backed by process-local event collections.
InMemoryRunner: Runner wired with in-memory session and artifact services.
InMemorySessionService: Session service backed by process-local maps.
InMemorySkillBuilder: Builder for one in-memory skill.
InMemorySkillData: Complete in-memory payload for one skill source entry.
InMemorySkillSource: In-memory SkillSource implementation.
InMemorySkillSourceBuilder: Builder for InMemorySkillSource.
InMemoryTelemetryService: In-memory telemetry implementation for local development and tests.
InvocationContext: Mutable invocation context shared across agent execution steps.
InvocationPlan: One invocation strategy used to call flexible user functions.
JoinNode: Node that passes through aggregated predecessor inputs.
ListSessionsResponse: Response model for session list operations.
ListSkillsTool: Tool to list all locally available skills.
LiveActivityEnd: Marker indicating the end of a live activity segment.
LiveActivityStart: Marker indicating the start of a live activity segment.
LiveConnectConfig: Realtime connection configuration for live model sessions.
LiveRequest: A single queued live request payload.
LiveRequestQueue: In-memory FIFO queue for live requests.
LiveSkillGcsStore: Web stub for live GCS-backed skill storage.
LlmAgent: Core LLM agent that drives model and tool orchestration flows.
LlmRequest: Full model request payload used by LLM adapters.
LlmResponse: Normalized model response emitted by LLM adapters.
LlmToolConfig: Tool-level configuration wrapper for model requests.
LoadMcpResourceTool: Tool that injects MCP resource contents into pending LLM requests.
LoadSkillResourceTool: Tool to load a resource from a skill bundle.
LoadSkillTool: Tool to load a skill's main instructions.
LocalSkillSource: Filesystem-backed SkillSource, unsupported on Web.
LoopAgent: Workflow agent that iterates through sub-agents repeatedly.
LoopAgentState: Serialized resumability state for LoopAgent.
McpBaseTool: Lightweight MCP tool descriptor used by McpTool.
McpResourceContent: Content returned by resources/read.
McpSessionManager: Central registry and transport manager for MCP sessions.
McpTool: Auth-aware MCP tool wrapper that delegates calls through McpSessionManager.
McpToolset: Toolset that exposes tools and resources from one MCP connection.
MemoryEntry: One memory item with content and optional metadata.
Node: Node designed for subclassing.
NodeInfo: Workflow node metadata attached to an Event.
NodePathBuilder: Represents a path to a node in a hierarchical workflow.
NodeState: Mutable state for one workflow node.
NoopAdkMetricsRecorder: Metrics recorder that intentionally drops all samples.
ParallelAgent: Workflow agent that runs sub-agents concurrently.
ParallelWorker: Node that runs a wrapped child node for every input item in parallel.
Part: One multimodal content part within a Content turn.
PluginManager: Coordinates plugin registration and callback dispatching.
ReplaySequenceBarrier: Chronological barrier used to coordinate deterministic replay ordering.
RequestInput: Data used to request additional user input from a workflow or agent.
RequestInputTool: Long-running tool that requests human input before execution continues.
Resources: L3 skill resources loaded from references/assets/scripts directories.
ResumabilityConfig: Resumability settings for an app.
RunConfig: Configuration object controlling one agent run behavior.
Runner: Coordinates session lifecycle, plugins, and agent execution.
RunSkillScriptTool: Tool to execute scripts from a skill bundle.
Script: Wrapper for script content.
SearchMemoryResponse: Response model returned by memory search operations.
SearchSkillsTool: Tool to search a configured skill registry for relevant skills.
SequentialAgent: Workflow agent that runs sub-agents one-by-one.
SequentialAgentState: Serialized resumability state for SequentialAgent.
Session: A conversational session containing state and event history.
SingleTurnAgentTool: Tool adapter for a sub-agent configured with mode: 'single_turn'.
Skill: Complete skill representation with metadata, instructions, and resources.
SkillDescriptor: Minimal shape shared by all skill descriptors.
SkillGcsStore: Web stub for GCS-backed skill storage.
SkillRegistry: In-memory registry for named Skill objects.
SkillSource: Asynchronous source interface for skill metadata, instructions, resources.
SkillToolset: Toolset exposing skill-discovery, loading, and script-execution tools.
State: Map-like state container with change tracking.
StdioConnectionParams: Process launch options for an MCP stdio server.
StreamableHTTPConnectionParams: Connection options for an MCP server that supports Streamable HTTP.
StructuredEvent: Base class for typed structured events.
TaskAgentTool: Tool adapter for a sub-agent configured with mode: 'task'.
TaskRequest: A request to delegate work to a task agent.
TaskResult: Result returned by a task agent after completion.
TelemetryLogRecord: One log entry recorded under a telemetry span.
TelemetrySpan: Span model representing one traced operation.
ThoughtEvent: Reasoning text emitted by the model.
ToolCallEvent: Request to execute a tool.
ToolConfirmationEvent: Request for user confirmation of tool calls.
ToolDeclaration: Tool declaration payload sent to model backends.
ToolNode: Node that wraps an ADK BaseTool.
ToolResultEvent: Result returned by a tool.
ToolThreadPoolConfig: Thread-pool configuration for concurrent tool execution.
UrlContextTool: Built-in URL context retrieval tool for supported Gemini models.
VertexAiRagResource: Vertex AI RAG resource reference.
VertexAiRagRetrieval: Retrieval tool that integrates with Vertex AI RAG.
VertexAiRagStore: Store-level RAG query configuration used by VertexAiRagRetrieval.
VertexAiSearchConfig: Retrieval configuration used by VertexAiSearchTool.
VertexAiSearchDataStoreSpec: One Vertex AI Search data store reference.
VertexAiSearchTool: Built-in retrieval tool wrapper for Vertex AI Search.
VertexRagRetrievalTool: Server-side Vertex AI RAG retrieval tool.
Workflow: Node-based workflow agent.
WorkflowContext: Shared runtime context passed to workflow nodes.
WorkflowResult: Execution result from Workflow.runWorkflow.

Enums

EventType: Categories emitted by toStructuredEvents.
FunctionCallingConfigMode: Function-calling modes recognized by model providers.
NodeStatus: Runtime status for a workflow node.
StreamingMode: Streaming transport mode used by a run.
ToolErrorType: Semantic tool error types aligned with OpenTelemetry error attributes.
ToolExecutionMode: Execution mode used when a model requests multiple tools in one turn.

Constants

DEFAULT_ROUTE → const String: Default route value used by routed workflow edges.
defaultSkillSystemInstruction → const String: Default system instruction injected when skill tools are enabled.
finishTaskSuccessResult → const String: Success result returned by FinishTaskTool.run.
finishTaskToolName → const String: Function name used by task agents to signal completion.
genAiAgentInvocationDurationMetric → const String: Metric name for agent invocation durations.
genAiAgentRequestSizeMetric → const String: Metric name for agent request sizes.
genAiAgentResponseSizeMetric → const String: Metric name for agent response sizes.
genAiAgentWorkflowStepsMetric → const String: Metric name for agent workflow step counts.
genAiToolExecutionDurationMetric → const String: Metric name for tool execution durations.
genAiToolRequestSizeMetric → const String: Metric name for tool request sizes.
genAiToolResponseSizeMetric → const String: Metric name for tool response sizes.
START → const String: Sentinel node name used for workflow start edges.

Properties

adkMetricsRecorder → AdkMetricsRecorder: The process-wide metrics recorder used by ADK helpers.
no setter
requestInput → RequestInputTool: Default request-input tool instance.
final
urlContext → UrlContextTool: Shared URL context tool instance.
final

Functions

buildNode(Object nodeLike, {String? name, String description = '', List<String>? dependsOn, bool? rerunOnResume, bool? waitForOutput, RetryConfig? retryConfig, Duration? timeout}) → BaseNode: Converts a workflow-compatible object into a BaseNode.
contentSizeBytes(Content? content) → int: Returns the UTF-8 and inline-data byte size of content.
createRequestInputEvent(RequestInput requestInput, {String invocationId = '', String author = '', String? role}) → Event: Creates a long-running function-call event for requestInput.
createRequestInputResponse(String interruptId, Map<String, dynamic> response) → Part: Creates a function-response part for a request-input interrupt.
eventActionsFromJson(Map<String, Object?> json) → EventActions: Deserializes json into EventActions.
eventActionsToJson(EventActions actions) → Map<String, Object?>: Serializes actions into a JSON-compatible map.
eventNodeInfoFromJson(Map<String, Object?> json) → NodeInfo: Deserializes json into workflow NodeInfo.
eventNodeInfoToJson(NodeInfo nodeInfo, {bool snakeCase = false}) → Map<String, Object?>: Serializes nodeInfo into a JSON-compatible map.
formatSkillsAsXml(List<SkillDescriptor> skills) → String: Renders a skill summary list as XML text for model prompts.
getRequestInputInterruptIds(Event event) → List<String>: Returns request-input interrupt ids contained in event.
getTime() → double: Returns the current time in seconds since the Unix epoch.
getUtcNow() → DateTime: Returns the current UTC time derived from getTime.
hasRequestInputFunctionCall(Event event) → bool: Whether event contains a request-input function call.
listSkillsInDir(String skillsBasePath) → Map<String, Frontmatter>: Returns no directory-backed skills on Web.
listSkillsInGcsDir(String bucketName, {String skillsBasePath = '', Object? storageStore}) → Future<Map<String, Frontmatter>>: Throws because GCS-backed skill loading is unsupported on Web.
loadSkillFromDir(String skillDirPath) → Skill: Throws because directory-based skill loading is unsupported on Web.
loadSkillFromGcsDir(String bucketName, String skillId, {String skillsBasePath = '', Object? storageStore}) → Future<Skill>: Throws because GCS-backed skill loading is unsupported on Web.
loadSkillFromZipBytes(List<int> zipBytes) → Skill: Loads a Skill from ZIP bytes containing a root-level SKILL.md.
newUuid() → String: Returns a new unique identifier string.
node(WorkflowFunction function, {required String name, String description = '', List<String>? dependsOn, bool rerunOnResume = false, RetryConfig? retryConfig, Duration? timeout}) → FunctionNode: Convenience wrapper matching Python's node(...) helper.
parallelWorker(Object nodeLike, {int? maxConcurrency, RetryConfig? retryConfig, Duration? timeout}) → ParallelWorker: Wraps nodeLike in a ParallelWorker.
readSkillProperties(String skillDirPath) → Frontmatter: Throws because skill-property file loading is unsupported on Web.
recordAgentInvocationDuration(String agentName, Duration duration, {Object? error, AdkMetricsRecorder? recorder}) → void: Records the duration of an agent invocation.
recordAgentRequestSize(String agentName, Content? userContent, {AdkMetricsRecorder? recorder}) → void: Records the byte size of the user content sent to an agent.
recordAgentResponseSize(String agentName, List<Event>? events, {AdkMetricsRecorder? recorder}) → void: Records the byte size of the last response content authored by an agent.
recordAgentWorkflowSteps(String agentName, List<Event> events, {AdkMetricsRecorder? recorder}) → void: Records the number of events authored by an agent.
recordToolExecutionDuration(String toolName, String agentName, Duration duration, {Object? error, AdkMetricsRecorder? recorder}) → void: Records the duration of a tool execution.
recordToolRequestSize(String toolName, String agentName, Map<String, Object?> functionArgs, {AdkMetricsRecorder? recorder}) → void: Records the byte size of string-valued tool arguments.
recordToolResponseSize(String toolName, String agentName, Event? responseEvent, {AdkMetricsRecorder? recorder}) → void: Records the byte size of a tool response event.
resetAdkMetricsRecorderForTest() → void: Restores the process-wide metrics recorder to a no-op recorder.
resetIdProvider() → void: Restores the default unique ID provider.
resetTimeProvider() → void: Restores the default system time provider.
setAdkMetricsRecorder(AdkMetricsRecorder recorder) → void: Sets the process-wide ADK metrics recorder.
setIdProvider(IdProvider provider) → void: Sets the provider used by newUuid.
setTimeProvider(TimeProvider provider) → void: Sets the provider used by getTime.
toolRequestSizeBytes(Map<String, Object?> functionArgs) → int: Returns the request byte size for string-valued tool arguments.
toStructuredEvents(Event event) → List<StructuredEvent>: Converts a raw Event into typed structured events.
validateAppName(String name) → void: Validates that name is a supported app identifier.
validateSkillDir(String skillDirPath) → List<String>: Reports unsupported validation mode on Web.

Typedefs

AfterModelCallback = FutureOr<LlmResponse?> Function(CallbackContext callbackContext, LlmResponse llmResponse): Callback invoked after model generation.
AfterToolCallback = FutureOr<Map<String, dynamic>?> Function(BaseTool tool, Map<String, dynamic> args, ToolContext toolContext, Map<String, dynamic> toolResponse): Callback invoked after a tool call succeeds.
Agent = LlmAgent: Alias for LlmAgent kept for API parity.
AgentLifecycleCallback = FutureOr<Content?> Function(CallbackContext callbackContext): Callback signature for agent lifecycle hooks.
BeforeModelCallback = FutureOr<LlmResponse?> Function(CallbackContext callbackContext, LlmRequest llmRequest): Callback invoked before model generation.
BeforeToolCallback = FutureOr<Map<String, dynamic>?> Function(BaseTool tool, Map<String, dynamic> args, ToolContext toolContext): Callback invoked before a tool call executes.
CallbackContext = Context: Mutable agent context exposed to callbacks.
ConfirmationPredicate = FutureOr<bool> Function(Map<String, dynamic> args): Callback signature for dynamic confirmation requirements.
GeminiGenerateHook = Stream<LlmResponse> Function(LlmRequest request, bool stream): Hook for overriding Gemini generation behavior.
GeminiInteractionsInvoker = Stream<LlmResponse> Function(LlmRequest request, {required bool stream}): Invoker hook for Gemini interactions API generation.
GeminiLiveSessionFactory = GeminiLiveSession Function(LlmRequest request): Factory that creates live Gemini session transports.
GoogleLlm = Gemini: Backward-compatible alias for Gemini.
IdProvider = String Function()
InstructionProvider = FutureOr<String> Function(ReadonlyContext context): Dynamic instruction provider callback.
JsonMap = Map<String, dynamic>: JSON-compatible map type used across model and tool payloads.
McpConnectionParams = Object: Union type alias for supported MCP transport connection parameters.
McpHeaderProvider = Map<String, String> Function(ReadonlyContext readonlyContext): Header-builder callback applied to MCP tool calls in this toolset.
McpSamplingCallback = FutureOr<Object?> Function(List<Map<String, Object?>> messages, {Map<String, Object?>? params}): Callback invoked when an MCP server requests client-side sampling.
McpToolExecutor = FutureOr<Object?> Function(Map<String, dynamic> args, {Map<String, String>? headers}): Callback signature for custom local MCP tool executors.
McpToolHeaderProvider = Map<String, String> Function(ReadonlyContext readonlyContext): Header-builder callback used before making MCP tool HTTP calls.
OnModelErrorCallback = FutureOr<LlmResponse?> Function(CallbackContext callbackContext, LlmRequest llmRequest, Exception error): Callback invoked when model generation throws.
OnToolErrorCallback = FutureOr<Map<String, dynamic>?> Function(BaseTool tool, Map<String, dynamic> args, ToolContext toolContext, Exception error): Callback invoked when a tool call throws.
SkillMetadataValue = Object?: Metadata value allowed in skill frontmatter.
SkillResourceData = Object: Resource payload allowed in skill references/assets.
StructuredErrorEvent = ErrorEvent: Alias for callers that prefer an unambiguous Dart-specific name.
TimeProvider = double Function()
ToolContext = Context: Mutable tool execution context alias used by tool interfaces.
VertexAiRagQueryHandler = Future<List<String>> Function({List<String>? ragCorpora, List<VertexAiRagResource>? ragResources, int? similarityTopK, required String text, double? vectorDistanceThreshold}): Callback signature for live Vertex AI RAG querying.
WorkflowFunction = FutureOr<Object?> Function(WorkflowContext context, Object? nodeInput): Function callback signature for FunctionNode.

Exceptions / Errors

AdkAbortException: Exception thrown when an operation observes an aborted ADK signal.
AlreadyExistsError: Exception indicating that a resource already exists.
InputValidationError: Exception indicating invalid user or API input.
LlmCallsLimitExceededError: Error thrown when LLM call count exceeds the configured limit.
McpHttpStatusException: Thrown when an HTTP request returns a non-success status code.
McpJsonRpcException: Thrown when a JSON-RPC response contains an error object.
NodeTimeoutError: Error thrown when a node exceeds its timeout.
NotFoundError: Exception indicating missing resources.
PluginManagerException: Exception thrown for plugin manager callback/close failures.
ResourceExhaustedModelException: Exception thrown when model capacity is exhausted.
SessionNotFoundError: Error thrown when a requested session cannot be found.
SkillSourceException: Exception used by SkillSource implementations.
ToolExecutionError: Exception raised when a tool fails with a semantic error classification.

Classes

Enums

Constants

Properties

Functions

Typedefs

Exceptions / Errors

adk_dart package

adk_core library