antigravity

Release v0.9.2 - Fix RetryConfig proto serialization, subagent tool n…

• Fix RetryConfig Serialization:
  • ModelAPIRetryConfig.toMap()/toJson() (and any RetryConfig containing one) threw MapperException: Unknown type Duration and emitted a spurious initial_sleep_duration key that localharness rejects. The initialSleepDuration Duration is a convenience argument, not a wire field, but was being generated as one.
  • Added ModelAPIRetryConfig.raw() as the @MappableConstructor(), so only the integer initialSleepDurationMs is serialized. The unnamed constructor still accepts either initialSleepDurationMs or initialSleepDuration.
• Fix Subagent Tool Protobuf Naming:
  • Corrected harness_side_tools field names for subagents in LocalConnectionStrategy ("file_edit" and "write_to_file", matching localharness.proto), resolving unknown field "edit_file" errors when launching static subagents.
• CI & Example Automation:
  • Added non-interactive terminal fallback (stdin.hasTerminal) to example/getting_started/human_in_the_loop.dart to prevent test runs from blocking on headless standard input.

• Critical Bug Fixes for Local Harness Execution:
  • Trajectory Idle State Recognition: Added support for STATE_FULLY_IDLE (and FULLY_IDLE) in LocalConnection trajectory state updates, properly emitting the idle_sentinel step and preventing agent.chat() / response.text() from hanging at turn completion.
  • Tool Authorization Handshake: Added handler for LIFECYCLE_HOOK_PRE_TOOL / PRE_TOOL in HookRouter, returning valid pre_tool_result decisions (ALLOW / DENY) to the harness and resolving tool execution deadlocks.
  • Wire Path Normalization: Added wire-format URI normalization (file:///, cns://) for tool call arguments in HookRouter to support path-based safety policies and canonical workspace containment.

Release v0.9.0 - Sync Python SDK v0.1.11 features, budget limits, sub…

• Sync with Python SDK v0.1.11:
  • Default Model Upgrade to Gemini 3.7 Flash: Updated default generative text model in lib/src/models.dart to gemini-3.7-flash.
  • Session Budget Enforcement (BudgetConfig) & Stop Reasons (StopReason):
    • Introduced BudgetConfig to configure session-level caps: maxModelCalls, maxToolCalls, maxInputTokens, maxOutputTokens, maxTotalTokens.
    • Added StopReason enum (unspecified, maxModelCallsExceeded, maxToolCallsExceeded, maxInputTokensExceeded, maxOutputTokensExceeded, maxTotalTokensExceeded, quotaExhausted).
    • Exposed ChatResponse.stopReason, ChatResponse.usage, and Conversation.lastTurnStopReason.
  • Vertex AI Express Mode (apiKey):
    • Added apiKey support to VertexEndpoint for Vertex AI Express Mode authentication.
    • Implemented mutual exclusivity validation: rejects configurations passing both apiKey (Express Mode) and project/location (Standard Mode).
  • Agent Behavior Standard (AgentBehavior):
    • Renamed AgentMode enum to AgentBehavior (autonomous, interactive), mapping to AGENT_BEHAVIOR_AUTONOMOUS and AGENT_BEHAVIOR_INTERACTIVE protobuf values.
    • Provided typedef AgentMode = AgentBehavior and constructor/getter aliases for backward compatibility.
  • Hierarchical & Nested Subagent Controls:
    • Added maxSubagentDepth and allowedSubagents allowlist configuration on CapabilitiesConfig and SubagentCapabilities.
    • Enabled BuiltinTools.startSubagent on custom subagents when configured in their capability toolsets.
    • Validated that maxSubagentDepth and allowedSubagents cannot be set when subagents are disabled.
    • Added subagent allowlist validation on BaseLocalAgentConfig ensuring all referenced subagent names exist in the configured subagents list.
  • Step Metadata & Image Artifact Paths:
    • Added parentTrajectoryId and depth fields to Step and updated Step.fromMap parser.
    • Added outputPath to GenerateImageResult and updated toString() to return outputPath when non-empty.
  • Policy Enhancements:
    • Made handler optional in askUser policy builder for host platform delegation compatibility.
  • New Example & Documentation:
    • Added example/getting_started/budget_limits.dart demonstrating all 5 budget limits and stop reason inspection.
    • Updated example/getting_started/subagents.dart showcasing dynamic self-delegation, static subagents, and nested subagent hierarchies.
  • SDK Version Bump & Dependency Synchronization:
    • Updated package version to 0.9.0 and MCP implementation version to 0.9.0.

• Sync with Python SDK v0.1.10 & Dart-Idiomatic Enhancements:
  • Breaking API Change: Conversation.lastTurnUsage signature updated to UsageMetadata? (matching Python SDK v0.1.10) to return null when no token usage was recorded during a turn.
  • Gemini Prioritized Inference (ServiceTier): Introduced ServiceTier enum (standard, priority, flex) and support in GeminiModelOptions and UsageMetadata to configure high-priority model execution with automated fallback.
  • Live Token Usage Updates: Added support for live UsageUpdate event streaming over WebSockets, accumulating per-trajectory token usage live during execution. Added subtraction (-) operator support to UsageMetadata.
  • Agent Execution Modes (AgentMode): Introduced AgentMode enum (autonomous, interactive) in CapabilitiesConfig and SubagentCapabilities with interactive tool validation warnings (BuiltinTools.askQuestion).
  • Interactive CLI Spinner: Updated interactive CLI step spinner formatting (formatStepSpinnerMessage) to display all active tool names during concurrent tool calls (e.g. Running tools 'tool_a', 'tool_b'...).
  • Tool Call Correlation ID: Added callId correlation support across ToolCall, ToolResult, ToolExecutionException, and lifecycle hook payloads.
  • Flexible Context-Aware Lifecycle Hooks: Added .stateless() factory constructors to FunctionInspectHook, FunctionDecideHook, and FunctionTransformHook to seamlessly support both (data) and (context, data) callback signatures.
  • ActionCompaction Handling: Emits context window compaction events over WebSockets and dispatches registered OnCompactionHook handlers.
  • Standardized System Instructions Strategy: Plain string instructions default to an appended system instruction strategy, while CustomSystemInstructions completely replaces built-in instructions.
  • LiteRT Token Output Limit: Increased LiteRT default maxOutputTokens from 8192 to 16,384 tokens to prevent truncation during complex generation.

Release v0.7.0 - Sync Python SDK v0.1.9 updates and Dart-idiomatic SD…

• Sync with Python SDK v0.1.9 & Dart-Idiomatic Enhancements:
  • Model-Call Retry & Backoff Configuration: Introduced RetryConfig, ModelAPIRetryConfig, and ModelOutputRetryConfig with RetryConfig.benchmark() preset for controlling exponential backoff and output retry behavior across LocalAgentConfig, LocalOpenAIAgentConfig, and LiteRTAgentConfig.
  • Strongly-Typed Duration & Jitter Ratio Validation: Added Duration support (initialSleepDuration, serverTimeout) to ModelAPIRetryConfig and McpServerConfig. Validates jitterRange as a ratio between 0.0 and 1.0.
  • Strongly-Typed Logging with package:logging: DebugConfig accepts strongly-typed Level objects (level: Level.WARNING) or string names, validating levels and applying process-safe logger updates on Agent.start().
  • Uri & Uint8List Media Buffers: Added Uri getters (uri, baseUri) on McpStreamableHttpServer and ModelEndpoint, plus McpStreamableHttpServer.fromUri() factory constructor. Added Uint8List get bytes getter on MediaContent and File object support in fromFile().
  • Step & Turn Extensions: Added extension getters (isUserTurn, isModelTurn, isFinished, isDone) on Step, StepSource, and StepStatus.
  • LiteRT Warm-up Timeout Scaling: Dynamically scales LiteRT engine warm-up timeouts based on context size (maxContextTokens).
  • Expanded Audio Payload MIME Support: Added support for audio/x-wav, audio/wave, audio/vnd.wave, audio/mp4, audio/webm, and additional audio formats.
  • Prompt Validation: Added strict validation in Conversation.chat() to reject null, empty string, whitespace-only, or empty content prompts with AntigravityValidationException.
  • Hybrid Exception Hierarchy: AntigravityValidationException implements both Exception and ArgumentError for backwards compatibility.
  • SDK Alignment: Bumped SDK version to 0.7.0 aligned with Python SDK v0.1.9 release.

Release v0.6.0: Sync Python SDK v0.1.8 updates

• Sync with Python SDK v0.1.8:
  • Default Model Upgrade to Gemini 3.6 Flash: Updated default generative text model in lib/src/models.dart to gemini-3.6-flash.
  • Prompt Sanitization: Strips null bytes (\x00) and dangerous control characters (DEL, BEL, C1) from incoming string prompts at the wire boundary.
  • Tool Execution Exception: Introduced ToolExecutionException carrying message, toolName, and serverName metadata when tool execution fails, with updated hook routing in HookRouter.
  • Subagent Custom & Templated System Instructions: Expanded SubagentConfig system instructions to support CustomSystemInstructions and TemplatedSystemInstructions.
  • SDK Version Bump & Alignment: Bumping SDK version to 0.6.0 aligned with Python SDK v0.1.8 release updates.

• Sync with Python SDK v0.1.7:
  • Hierarchical Thread-Safe State Management: Introduced the StateStore class for hierarchical context state sharing and lock/reentrancy support. Retargeted HookContext and ToolContext to inherit from StateStore.
  • Session Continuation Modes: Implemented SessionContinuationMode to specify continuation behavior (RESUME, CREATE_OR_RESUME, CREATE_ONLY, SESSION_CONTINUATION_MODE_UNSPECIFIED) with validation checks on AgentConfig.
  • Dynamic Environment Hydration: Hydrates Vertex project and location parameters automatically from standard GOOGLE_CLOUD_PROJECT and GOOGLE_CLOUD_LOCATION environment variables when not explicitly passed, and enables Vertex mode automatically when GOOGLE_GENAI_USE_VERTEXAI or GOOGLE_GENAI_USE_ENTERPRISE is set.
  • Websocket Retry Backoff & DNS Resolution: Refactored websocket connections in LocalConnectionStrategy to retry on both localhost and 127.0.0.1 to ensure reliable connections in containerized sandboxes.
  • Spinner Pause/Resume: Enhanced CLI Spinner with pause and resume behaviors, and integrated them into user confirmation/question hooks to prevent prompt clobbering.
  • Custom Tool Execution Support: Added custom_tool parsing support in Step.fromMap and filtered out tool calls from StepUpdate if they are local custom tools to prevent client duplicate events.
  • Addition Operator for Token Usage: Implemented + addition operator on UsageMetadata for clean accumulation of token usage.
  • Updated default image generation model to 'gemini-3.1-flash-lite-image'.
  • Added ThinkingLevel.extraHigh ('extra_high') level support.

chore: bump version to 0.4.1 and apply post-release bug fixes & docum…

• Bug Fixes & Adjustments:
  • Restored default workspaces (current working directory) and capabilities configuration on BaseLocalAgentConfig to prevent silent sandboxing regressions.
  • Fixed client handshake protocol to announce the correct client version instead of a stale one.
  • Added missing read_url_content tool field mapping in Step.fromMap parser.
  • Resolved HttpClient socket/connection leak in LiteRT loopback server health checks.
  • Documented public configuration fields on LocalOpenAIAgentConfig and LiteRTAgentConfig.

feat: sync Python SDK 0.1.6 updates & add localharness auto-upgrades

• Local Inference & OpenAI Endpoint Support:
  • Added support for local Gemma execution using LiteRT via LiteRTAgentConfig and LiteRTConnectionStrategy (which manages a python loopback HTTP server).
  • Added support for any OpenAI-compatible completions API (e.g. Ollama, LM Studio) via LocalOpenAIAgentConfig and LocalOpenAIConnectionStrategy.
• Multimodal Tool Outputs:
  • Enabled custom tools to return media assets (images, audio, video, documents) directly via a single tool response without needing separate follow-up turns. Media is extracted and sent as supplemental_media in tool_response.
• Model Context Protocol (MCP):
  • Decoupled tool calls and safety policy engines from legacy mcp_ string prefix synthesis, utilizing explicit serverName attributes in tool evaluation.
• Alias Deserialization Support:
  • Handled results/entries alias in ListDirectoryResult and output/combined_output alias in RunCommandResult.
• Automated Binary Discovery Updates:
  • Added version checking logic to verify if the cached localharness binary is out of date.
  • Automatically triggers a re-download/upgrade if the cached binary version is older than the SDK's default version (0.1.6).

chore: bump version to 0.3.1 and update CHANGELOG.md

• Fix Tool Error Hook Dispatching:

  • Resolved a bug where OnToolErrorHook was never dispatched for either client-side custom tool failures or harness-side built-in tool failures.
  • Implemented client-side tool error dispatching in local_connection.dart with support for recovery values.
  • Added routing for LIFECYCLE_HOOK_ON_TOOL_ERROR and ON_TOOL_ERROR hook requests in hook_router.dart and covered it with a unit test.

• Dart 3 Modernization:

  • Refactored core classes Connection and ConnectionStrategy to use Dart 3 abstract interface class modifiers.
  • Marked MediaContent as a sealed class to restrict subclass hierarchy and support compiler-level exhaustiveness checks.
  • Modernized registerHook, allow, deny, and askUser to leverage case type pattern matching inside switch statements.
  • Refactored policy bucket index resolution (_bucketIndex), file change triggers (onFileChange), shorthand model builder (_buildShorthandModels), and step state mappings (Step.fromMap) to use clean switch expressions and logical OR patterns.
  • Updated local connection classes to implement rather than extend the new interface classes.
  • Added conditional imports in lib/src/types/content.dart to support running the package on the web.

• Synchronize updates from Python SDK (v0.1.5):
  • Added new read_url_content builtin tool and its ReadUrlContentResult structured output.
  • Added StepType.thinking to enum representation for telemetry tracking.
  • Implemented the HookRouter to process WebSocket-based CallHookRequest turn and tool lifecycle hooks delegated by the harness.
  • Added support for telemetry PreStepHook and PostStepHook inside HookRunner and connection parser.
  • Implemented _StepTracker to prevent duplicate step hooks or non-linear state transitions.
  • Supported error and cancelled propagation in trajectory_state_update events.
  • Historical Step Absorption: Aligned startup handshake to await and parse pre-existing conversation history and usage metadata from initialize_conversation_response, populating them immediately in Conversation on start.

• Reintroduce generated files in package

• Model Configuration Overhaul: Replaced the monolithic GeminiConfig with a more flexible ModelTarget and polymorphic ModelEndpoint class hierarchy (GeminiAPIEndpoint, VertexEndpoint).
• Subagents Feature: Introduced SubagentConfig and SubagentCapabilities allowing definition and inclusion of subagents in the main agent's configuration.
• New Builtin Tools: Added SearchWeb tool and related SearchWebResult structured output.
• Enhanced Configuration: Support for environment variables (env) in McpStdioServer. Removed imageModel from CapabilitiesConfig to favor the new generic model targeting features.

0.1.3

• Added support for Vertex AI configuration (project, location, vertex options) in GeminiConfig.
• Refactored Model Context Protocol (MCP) server configurations to run connections natively on the Go-based harness, removing client-side McpBridge references.
• Added SlashCommand structure to input content primitives to support built-in planning flows.
• Integrated CLI Spinner into runInteractiveLoop with step-by-step progress tracking via agent.conversation.receiveSteps().
• Enhanced policy engine with a 9-level priority model supporting prefix wildcards and custom MCP policy validators.
• Added AntigravityCancelledException and AntigravityExecutionException for robust error handling.

0.0.1

• Initial release of the Antigravity Dart SDK.