adk_dart

🚀 What's New in v2026.8.17

🌟 Core Framework & Upstream Sync (ADK v2.7.0 & JS v1.6.0 Parity)

• LlmCapabilities (lib/src/models/capabilities.dart): Added standardized capability negotiation system (outputSchemaAndTools, etc.) across BaseLlm and Gemini.
• GetUserChoiceTool (lib/src/tools/get_user_choice_tool.dart): Added user choice interruption and multi-turn waiting helper (getUserChoice).
• FunctionTool Parameter Feedback: Return self-correcting parameter error guidance for missing mandatory args to prompt the LLM to provide required parameters.
• Multimodal Tool Results Extraction: Automatic separation of inline binary parts into FunctionResponse.parts to prevent JSON serialization issues.
• Orphaned FunctionResponse Pruning: Automatic pruning in contents.dart to prevent Gemini 400 Bad Request errors when function calls are compacted/removed.
• Memory Namespace Isolation: Tuple record keys (appName, userId) in InMemoryMemoryService to prevent delimiter collisions when identifiers contain slashes.
• Streaming Thought Signature Preservation: Seamless thoughtSignature preservation across streaming text chunks.
• StaleSessionError (lib/src/errors/stale_session_error.dart): Optimistic concurrency exception handling in SQLite & database session services.
• adk-web & Telemetry Endpoint: Full synchronization of compiled UI bundle assets (lib/src/cli/browser/) + CSRF-protected /config/telemetry endpoint.

📱 flutter_adk & Sub-packages

• flutter_adk (v2026.8.17): Exported Web-safe parity symbols (LlmCapabilities, GetUserChoiceTool, StaleSessionError, SessionNotFoundError, ReflectAndRetryModelPlugin, ReflectAndRetryToolPlugin) + upgraded web: ^1.1.1 and plugin_platform_interface: ^2.1.8.
• adk_litertlm (v2026.8.17): Upgraded litertlm: ^0.0.13 and aligned runtime interface with new conversation methods.
• adk_mcp (v2026.8.17): Synchronized client protocol primitives and upgraded dependencies.
• adk (v2026.8.17): Package alias updated to 2026.8.17.

📦 Published Packages on pub.dev

• adk_dart 2026.8.17
• adk 2026.8.17
• adk_mcp 2026.8.17
• flutter_adk 2026.8.17
• adk_litertlm 2026.8.17

• Synced latest core features & bug fixes from adk-python v2.7.0, adk-js v1.6.0, adk-go v2.2.0, and adk-kotlin v0.7.0:
  • Added LlmCapabilities (lib/src/models/capabilities.dart) model feature capability reporting system and updated BaseLlm.capabilities & Gemini.capabilities.
  • Added GetUserChoiceTool (lib/src/tools/get_user_choice_tool.dart) with getUserChoice callable helper for user choice prompts.
  • Enhanced FunctionTool with self-correction feedback returned to the model when mandatory required parameters are missing.
  • Added 20MB file size guard (maxArtifactSizeBytes = 20 * 1024 * 1024) to SaveFilesAsArtifactsPlugin to protect against oversized artifact payloads.
  • Supported multimodal media extraction (FunctionResponse.parts) from tool execution results in functions.dart.
  • Added automatic orphaned function_response pruning (_dropOrphanedFunctionResponses) in ContentsLlmRequestProcessor (contents.dart).
  • Improved InMemoryMemoryService with tuple record keys ((appName, userId)) to prevent delimiter collisions when identifiers contain slashes.
  • Enhanced StreamingResponseAggregator to preserve thoughtSignature across streamed text chunk runs without premature buffer flushing.
  • Aligned Gemini EAP model detection regex in model_name_utils.dart (r'^gemini-(?:[a-z0-9_]+(?:-[a-z0-9_]+)*-)?early-exp\d*$').
  • Added StaleSessionError (lib/src/errors/stale_session_error.dart) for optimistic concurrency write conflicts in database and SQLite session services.
  • Implemented Live API Voice Activity Detection (VAD) with VoiceActivity, VoiceActivityType, and explicitVadSignal in RunConfig and LiveConnectConfig.
  • Added FunctionResponseScheduling (whenIdle, silent, interrupt) and non-blocking background tool execution dispatching (activeNonBlockingToolTasks) in Live mode.
  • Implemented Workflow as Tool (WorkflowTool and Workflow.asTool()) allowing workflows to be invoked directly via function calling.
  • Added EventarcToolset and publishEventarcMessage for publishing structured CloudEvents 1.0 to Google Cloud Eventarc Advanced message buses.
  • Added DaytonaEnvironment execution environment for remote isolated cloud workspace sandboxes.
  • Implemented LlmAudioUserSimulator for multimodal speech/TTS conversation simulation and evaluation.
  • Added DatabaseSessionService.prepareTables() and BaseSessionService.prepareTables() for proactive table creation and migration.
  • Added create_data_agent tool in DataAgentToolset.
  • Added memory_id and allowed_topics support in VertexAiMemoryBankService.
  • Added Gemma 4 regex pattern matching (gemma-(3|4).*) in GemmaLlm.supportedModels().
  • Added comprehensive parity unit test suite in test/recent_upstream_parity_features_test.dart (1,517 total tests passing).

• Full Decoupling & Migration of packages/adk CLI Toolchain:
  • Physically migrated all CLI submodules (lib/src/cli/, lib/src/dev/, lib/src/cli/browser/, Angular Web UI static bundle) and 19 CLI/Dev test suites into packages/adk.
  • Fully purged all CLI/Dev/WebServer code and executables from adk_dart, establishing adk_dart as a 100% pure Core SDK library package.
  • Configured packages/adk as a standalone binary package publishable to pub.dev with direct dependency declarations.
• 100% CLI Parity with adk-python:
  • Created official parity matrix documentation in docs/architecture/cli_parity_matrix.md.
  • Implemented adk telemetry (enable, disable, status) subcommands for telemetry preference management.
  • Implemented adk test [folder] for executing agent test suites.
  • Added Dart-ecosystem diagnostic command adk doctor / adk diag displaying system, SDK, architecture, and path diagnostics.
• Examples Modernization & gemini-3.7-flash Upgrade:
  • Upgraded all 18 examples in adk_dart/examples and CLI starter templates to use gemini-3.7-flash by default.
  • Applied Dart 3.8+ dot-shorthands (.userText(...), .modelText(...), .user, .text(...)) across all example entrypoints.
• Quality Assurance:
  • 100% test pass rate across all workspace packages: 1,369 tests in adk_dart, 159 tests in packages/adk, 11 tests in flutter_adk, 12 tests in adk_mcp, 5 tests in adk_litertlm (1,556 tests total).
  • 0 analyzer warnings / errors across packages/adk, packages/flutter_adk, packages/adk_mcp, packages/adk_litertlm, and examples.

• Updated all README installation and dependency configuration sections across all ecosystem packages to 2026.8.17+2.

• Updated default model baseline across LlmAgent to gemini-3.7-flash.
• Exported LiteLlm in adk_core.dart for Web-safe local LLM (Ollama / LM Studio) integration without dart:io dependencies.

2026.7.30

• Bumped package release versions to 2026.7.30 for adk_dart, adk, adk_mcp, flutter_adk, and adk_litertlm.
• Added ReflectAndRetryModelPlugin for framework-level LLM error recovery, self-healing reflection tool calls (adkHandleModelError), and retry tracking.
• Added FunctionTool declaration caching (_cachedDeclaration) to eliminate redundant parameter schema rebuilds during agent turns.
• Standardized SessionNotFoundError handling across session services, agent runners, and dev web server.
• Updated MCP protocol support to official specification version 2026-07-28 (mcpLatestProtocolVersion = 2026-07-28).
• Added elicitationCallback support (elicitation/create and notifications/elicitation/complete) to McpToolset and McpSessionManager for client-side user input & OAuth challenge handling.
• Added AdkAgentMcpServer to expose any ADK BaseAgent as a compliant MCP server over tool interfaces.
• Implemented LoadWebPageTool and EnterpriseWebSearchTool (ported from adk-js) for web scraping with SSRF prevention and Vertex AI search grounding.

• Bumped package release versions to 2026.7.30 for adk_dart, adk, adk_mcp, flutter_adk, and adk_litertlm.
• Added ReflectAndRetryModelPlugin (lib/src/plugins/reflect_retry_model_plugin.dart) for framework-level LLM error recovery, self-healing reflection tool calls (adkHandleModelError), and retry tracking.
• Added FunctionTool declaration caching (_cachedDeclaration) to eliminate redundant parameter schema rebuilds during agent turns.
• Standardized SessionNotFoundError handling across session services, agent runners, and dev web server.
• Updated MCP protocol support to official specification version 2026-07-28 (mcpLatestProtocolVersion = '2026-07-28').
• Added elicitationCallback support (elicitation/create and notifications/elicitation/complete) to McpToolset and McpSessionManager for client-side user input & OAuth challenge handling.
• Added AdkAgentMcpServer (lib/src/tools/mcp_tool/agent_to_mcp.dart) to expose any ADK BaseAgent as a compliant MCP server over tool interfaces.
• Implemented LoadWebPageTool and EnterpriseWebSearchTool (ported from adk-js) for web scraping with SSRF prevention and Vertex AI search grounding.

• Bumped package release versions to 2026.7.24 for adk_dart, adk, adk_mcp, flutter_adk, and adk_litertlm.
• Synced latest core features & bug fixes from adk-python (be5828f3..07455ee6):
  • Added user labels (labels) support in RunConfig and automatic merging in BasicLlmRequestProcessor to propagate Vertex AI billing/attribution labels.
  • Guarded resumable invocation replay in BaseLlmFlow against partial streaming function-call SSE events to prevent replay loops.
  • Expanded AnthropicLlm finishReason mapping to support pause_turn (STOP) and refusal (SAFETY).
  • Fixed VertexAiSessionService.getSession to apply afterTimestamp server-side filtering alongside numRecentEvents.
  • Added opt-in finalResponseToolNames to BigQueryAgentAnalyticsPlugin to log final answer tool call payloads as AGENT_RESPONSE events.
• Added comprehensive unit test suite in test/python_upstream_updates_test.dart (1,105 total unit tests passing).
• Fixed nested workflow resume by yielding resolved outputs of resumed request-input nodes on resume instead of skipping them.
• Fixed AgentNode.run event loop to not misclassify endOfAgent and agentState checkpoint events as finalEvent.
• Enhanced LoadArtifactsTool with Gemini unsupported inline MIME type blocklist (SVG/XML image variants) to deliver them as text.
• Rebuilt and updated the bundled Web UI assets (adk-web) to the latest version to support nested navigation breadcrumbs, bidi streaming restarts, and usage token counts.
• Renamed the example folder to examples and reorganized all code examples into self-contained, user-friendly Dart project templates with individual README.md and dependency setup files in 4 languages (KO, EN, JA, ZH).
• Implemented a default HTTP completions invoker inside LiteLlm supporting streaming (SSE) and non-streaming requests to connect out-of-the-box with Ollama and LiteLLM servers.
• Added new local model examples: 08_local_llm_ollama_litellm (Ollama/LiteLLM integration via LiteLlm client) and 09_local_llm_litert (on-device Gemma inference via adk_litertlm sub-package).

What's New in 2026.7.11

• ADK 2.0 Workflows & Managed Agent (Interactions): Full support for the new ADK 2.0 features:
  • v2 Workflows: Declarative node graph scheduling, fan-out edges, state delta merging, and conditional routing via Workflow, BaseNode, JoinNode, etc.
  • Managed Agents: Integration with the GCP Managed Agents API (interactions.create endpoint) using ManagedAgent and RemoteMcpServer to leverage server-side execution.
• Header Propagation: Support for request-time header propagation and tracking headers across all operations.
• Static Analysis & Stability: Resolved static analysis warnings and improved streaming mode robustness in the interactions stream parser.

• Bumped package release versions to 2026.7.11 for adk_dart, adk, adk_mcp, flutter_adk, and adk_litertlm.
• Added ADK 2.0 Managed Agent support: ManagedAgent and RemoteMcpServer for server-side tool execution via the GCP Interactions API.
• Added environmentId propagation across LlmResponse, Event, session persistence, and telemetry tracing.
• Added isManagedAgent toggle to LlmRequest for grounding tool check bypass in managed workflows.
• Updated InteractionsProcessor.findPreviousInteractionState to return both previousInteractionId and environmentId for turn-to-turn chaining.
• Added isEnterpriseModeEnabled helper in env_utils.dart.
• Added ADK 2.0 Workflow & Managed Agent documentation to README (EN/KO).
• All 1,365 tests passing.

Changes

• Bumped release versions to 2026.6.6 for adk_dart, adk, adk_mcp, and flutter_adk.
• Updated the exported ADK runtime baseline to 2.2.0 to match the current adk-python reference baseline.
• Added broad workflow parity coverage, including workflow nodes, routed edges, dynamic run IDs, request-input helpers, retry/cancellation/concurrency handling, replay/resume/rehydration, nested workflow events, graph serialization, and status-aware graph DOT output.
• Expanded runtime event and conversion parity with structured events, node path building, A2A metadata preservation, Event routing helpers, and output-key callback visibility coverage.
• Added newer integration/tool surfaces, including GCP Skill Registry, BigQuery integration exports, environment tools, public skill tools, URL/load-web-page handling, Vertex RAG retrieval updates, and OpenAI labs adapter support.
• Hardened model, streaming, and tool-call behavior across Gemini schemas, Anthropic/LiteLLM/Apigee paths, REST streaming function continuations, streaming tool-call preservation, and cooperative abort handling.
• Updated adk_mcp non-IO stdio stub parity and hardened cancellation coverage.
• Refreshed flutter_adk example/documentation surface and consolidated repository documentation under docs/knowledge and docs/worklog.

Verification

• dart test passed: 1326 passed, 3 skipped.
• dart analyze lib test passed with existing info-level lint noise only.
• dart analyze / dart test passed for packages/adk.
• dart analyze / dart test passed for packages/adk_mcp.
• flutter analyze / flutter test passed for packages/flutter_adk.
• flutter pub get passed for packages/flutter_adk/example.

• Bumped package release versions to 2026.6.6 for adk_dart, adk, adk_mcp, and flutter_adk.
• Updated the exported ADK runtime baseline version to 2.2.0 to match the current local adk-python reference baseline.
• Added broad workflow parity coverage, including workflow nodes, routed edges, dynamic run IDs, request-input helpers, retry/cancellation/concurrency handling, terminal-output ownership, replay/resume/rehydration, nested workflow events, graph serialization, and status-aware graph DOT output.
• Expanded runtime event and conversion parity with structured events, node path building, A2A part metadata preservation, Event convenience routing, and output-key callback visibility coverage.
• Added and exposed newer integration/tool surfaces, including GCP Skill Registry, BigQuery integration exports, environment tools, public skill tools, URL/load-web-page handling, Vertex RAG retrieval updates, and OpenAI labs adapter support.
• Hardened model, streaming, and tool-call behavior across Gemini schemas, Anthropic/LiteLLM/Apigee paths, REST streaming function continuations, streaming tool-call preservation, and cooperative abort handling.
• Added telemetry/plugin parity updates, including metrics instrumentation, auto tracing plugin support, BigQuery analytics hardening, and save-files-as-artifacts behavior updates.
• Refreshed flutter_adk example and documentation surface and consolidated repository documentation under docs/knowledge and docs/worklog.

• Bumped package release versions to 2026.3.21 for adk_dart, adk, adk_mcp, and flutter_adk.
• Added latest parity coverage for session persistence and storage safety, including stricter stale-session detection, Vertex AI session metadata round-tripping, LiteLLM Anthropic thinking-block support, and Spanner database_role propagation.
• Added March 18-19 parity coverage for the experimental SpannerAdminToolset, including admin client/runtime factories, bootstrap wiring, and regression tests.
• Added environment_simulation rename-compatible exports and factory aliases on top of the existing agent simulator runtime.
• Hardened Anthropic tool-result serialization so arbitrary map/list payloads are emitted as JSON instead of lossy stringified Dart map text.
• Closed the latest four-day parity tail by adding MCP sampling callback/capability wiring across adk_mcp, McpSessionManager, and McpToolset, with HTTP integration coverage.
• Added multi-turn evaluation metric support for task success, trajectory quality, and tool-use quality, including registry/metric-info wiring and regression tests.
• Added Slack runner, Agent Registry integration, and GCP IAM connector compatibility surfaces to close remaining 3/19 runtime gaps.
• Added latest ref parity updates for A2A action-metadata round-tripping, toA2a(..., lifespan: ...), compaction safety around pending function calls, structured Discovery Engine datastore fallback, and feature-gated snake_case skill names.
• Added import-path compatibility exports for CrewAI and LangChain integrations under src/integrations/....

Full Changelog: 2026.3.1...2026.3.6

Highlights

• Fixed nested .adk/session.db path creation when app directory mappings resolve to absolute agent paths.
• Applied additional March adk-python parity updates across sessions, CLI agent loading, telemetry, LiteLLM parsing, and Bigtable metadata tooling.
• Finalized 2026.3.6 package alignment for adk_mcp, adk_dart, adk, and flutter_adk.
• Restored correct publish ignore boundaries so package-local pubspec.yaml, LICENSE, README.md, and CHANGELOG.md are visible during subpackage publish.

Included Since 2026.3.2+1

CLI and Runtime

• Added positional deploy target support:
  • adk deploy cloud_run
  • adk deploy agent_engine
  • adk deploy gke
• Implemented adk conformance test --mode live.
• Improved adk web / adk api_server project loading parity.
• Added /run auto session creation request handling.

Session and Storage

• Fixed local session DB path handling for absolute app directory mappings.
• Preserved temp: session state for later agents/tools within the same invocation without persisting it.

Parity Updates

• listAgents() now filters non-agent directories and respects single-app alias behavior.
• Telemetry now emits gen_ai.agent.version.
• LiteLLM now parses both reasoning_content and reasoning.
• Bigtable metadata tooling now includes cluster metadata support.

Package Versions

• adk_mcp: 2026.3.6
• adk_dart: 2026.3.6
• adk: 2026.3.6
• flutter_adk: 2026.3.6

Notes

• flutter_adk is a facade-alignment release on top of adk_dart 2026.3.6.
• Generated release notes below cover the full commit/PR range since 2026.3.2+1.

• Bumped package release versions to 2026.3.6 for adk_dart, adk, adk_mcp, and flutter_adk.
• Fixed nested .adk/session.db path creation when app directory mappings resolve to absolute agent paths.
• Applied the March Python parity batch covering temp-scoped session state visibility, listAgents() agent filtering, gen_ai.agent.version telemetry, LiteLLM reasoning parsing, and Bigtable cluster metadata tools.

• Bumped package release versions to 2026.3.2.
• Updated cross-package dependency alignment (adk_dart / adk / adk_mcp / flutter_adk) for the new release.

Full Changelog: 2026.3.1...2026.3.2+1

• Bumped package release versions using build metadata (2026.3.2+1) for adk_dart, adk, adk_mcp, and flutter_adk.
• Improved adk CLI parity for run/web/conformance flows to better match adk-python behavior.
• Fixed adk run . project path resolution when running from the agent directory.
• Updated conformance client/server wiring (/apps/.../sessions, /run_sse) and replay validation/report handling, including non-zero exit on test failures.
• Hardened local .adk path handling and project directory validation, with expanded CLI/web regression tests.

Full Changelog: 2026.2.28...2026.3.1

• Hardened A2A push callback delivery with a persistent SQLite queue, retry/backoff policy handling, startup/background drain, and dead-letter capture.
• Added A2A delivery reliability coverage in test/dev_web_server_test.dart, including persisted queue replay after restart.
• Added live network backend integration tests for PostgreSQL/MySQL session services (env-gated via ADK_TEST_POSTGRES_URL / ADK_TEST_MYSQL_URL).
• Added unified runtime adapter bootstrap APIs: configureToolRuntimeBootstrap(...) and resetToolRuntimeBootstrap(...).
• Added global default audio recognizer registration support via AudioTranscriber.registerDefaultRecognizer(...).
• Added bundled default runtime clients for BigQuery, Bigtable (admin/data), and Spanner (REST-based), removing mandatory client-factory injection for core paths.
• Added token resolution fallbacks for default cloud clients (explicit oauth token, environment variables, gcloud ADC, metadata server).
• Added built-in default Spanner embedder runtime via Vertex AI predict API (with env/full-model resource fallback handling and explicit fallback guidance).
• Added built-in default Toolbox HTTP delegate (/api/toolset/*, /api/tool/*/invoke) so ToolboxToolset can run without pre-registered delegate wiring.
• Added built-in live HTTP/auth providers in GcsArtifactService, removing mandatory provider injection for direct service usage.
• Ported telemetry parity for inference span/experimental semconv paths and added regression coverage.
• Added telemetry parity worklog and aligned package release versions to 2026.3.1.
• Updated parity tests to validate concrete default Bigtable/Spanner client availability.
• Refreshed runtime checklist and README feature matrix to reflect current implemented behavior.

Full Changelog: 0.1.1...2026.2.28

• Switched package versioning scheme to date-based YYYY.M.D.
• Expanded adk web / adk api_server parity surface with bundled /dev-ui serving and broader API route compatibility.
• Hardened MCP transport/session behavior through adk_mcp integration updates and related tests.
• Replaced deprecated mysql1 runtime dependency with mysql_client_plus in network session backend wiring.
• Added MySQL secure-fallback connect behavior for auth plugins that require TLS (for example caching_sha2_password).
• Added MySQL TLS hardening options (ssl_ca_file, ssl_cert_file/ssl_key_file, ssl_verify) with fail-fast validation for invalid TLS file configuration.
• Added runtime gap audit document for remaining non-functional/default-unwired features.

• Added package topics metadata on pub.dev to improve discoverability.

Full Changelog: https://github.com/adk-labs/adk_dart/commits/0.1.1

• Expanded Python parity coverage across evaluator workflows and remote code executor paths.
• Added parity ports for model contracts and toolset stacks, including Google API, Pub/Sub, and OpenAPI integrations.
• Added Spanner, Bigtable, and BigQuery client/toolset parity layers with vector store and metadata/query support.
• Added session schema/migration utilities and missing shared utility modules.

• Bootstrap adk_dart package as an ADK core runtime port (based on ref/adk-python).
• Added core agent/session/event/tool/model abstractions and execution flow.
• Added Runner / InMemoryRunner with async invocation path.
• Added function-call orchestration in LLM flow.
• Added baseline tests for event behavior, session service, and runner flow.