What's Changed

• fix: normalize nested models to JSON in generated setters by @pavelgj in #337
• docs: clarify Firestore project id discovery requirements by @pavelgj in #338

Full Changelog: genkit-v0.15.0...genkit-v0.15.1

Fixes

• normalize nested models to JSON in generated setters (#337)

Agents

This release introduces Agents: a first-class way to build stateful, multi-turn AI experiences in Dart. Agents wrap generate with durable sessions, tool calling, human-in-the-loop interrupts, sub-agent delegation, and a transport-agnostic client — so you can go from a one-shot prompt to a full conversational agent (running locally or served over HTTP) with a few lines of code.

Defining an agent is one call. It handles the model, tools, and session state for you:

import 'package:genkit/genkit.dart';
import 'package:genkit/io.dart';
import 'package:genkit_google_genai/genkit_google_genai.dart';

final ai = Genkit(
  plugins: [googleAI()],
  model: googleAI.gemini('gemini-flash-latest'),
);

final getWeather = ai.defineTool(
  name: 'getWeather',
  description: 'Get the current weather for a given location.',
  inputSchema: GetWeatherInput.$schema,
  outputSchema: GetWeatherOutput.$schema,
  fn: (input, _) async =>
      GetWeatherOutput(weather: 'Sunny in ${input.location}', temperature: '71F'),
);

final weatherAgent = ai.defineAgent(
  name: 'weatherAgent',
  system: 'You are an assistant helping with weather information. '
      'Use the getWeather tool.',
  tools: [getWeather],
  store: FileSessionStore('.sessions'), // durable, multi-turn state
);

Talk to it - streaming, multi-turn, with state carried automatically between turns:

final chat = weatherAgent.chat(sessionId: 'user-123');

// Stream the first turn.
final turn = chat.sendStream(text: 'What is the weather like in Tokyo?');
await for (final chunk in turn.stream) {
  stdout.write(chunk.text);
}

// Follow-up turn — prior context is threaded automatically.
final res = await chat.send(text: 'What about Paris?');
print(res.text);

Serve it over HTTP with genkit_shelf, and any client (Dart, JS, browser) can
drive it:

router.post('/api/weatherAgent', shelfHandler(weatherAgent.action));

// From a Dart CLI or Flutter/web app — no server-side deps required.
import 'package:genkit/client.dart';

final weather = remoteAgent(url: 'http://localhost:8080/api/weatherAgent');
final chat = weather.chat(sessionId: 'user-123');
await for (final chunk in chat.sendStream(text: 'Weather in Tokyo?').stream) {
  stdout.write(chunk.text);
}

What agents give you

• Durable sessions & snapshots - server- or client-managed state, with
  pluggable stores: InMemorySessionStore, FileSessionStore
  (package:genkit/io.dart), and FirestoreSessionStore (new
  genkit_google_cloud package). Load any past turn from a snapshot, or fork a
  conversation into a variant.
• Streaming multi-turn chat - send / sendStream, with session state
  threaded automatically across turns.
• Tools & human-in-the-loop interrupts - tools can pause for approval
  (ctx.interrupt(...)) and be resumed/restarted with an approval payload, so
  security-critical checks live inside the tool where the model can't bypass
  them.
• Typed custom state - define a stateSchema and mutate structured session
  state from tools; each change streams a live customPatch chunk to the client.
• Sub-agent delegation - the new agents() middleware auto-injects
  delegate_to_* tools, discovers sub-agent descriptions, and adds guard rails
  (maxDelegations, historyLength) for orchestrator/worker patterns.
• Prompt-file agents - definePromptAgent wires a .prompt (dotprompt) file
  into a multi-turn agent, customizable via promptInput.
• Custom & background agents - defineCustomAgent for multi-step flows with
  live status, plus detach + status polling + abort for long-running
  background work.
• Transport-agnostic client - remoteAgent from package:genkit/client.dart
  is browser-safe and wire-compatible with the Genkit client across SDKs
  (verified by a cross-SDK conformance suite).

See the full runnable showcase in testapps/agents - 11 demo pages covering
chat, interrupts, artifacts, background/detach, branching, task state, research,
delegation, and coding agents.

Agents changelog

• feat(agents): agent & session schema types (1/8) by @pavelgj in #310
• feat(agents): session & snapshot storage + JSON Patch (2/8) by @pavelgj in #311
• feat(agents): dart:io FileSessionStore (3/8) by @pavelgj in #312
• feat(agents): transport-agnostic client core (4/8) by @pavelgj in #313
• feat(agents): server-side agent runtime (5/8) by @pavelgj in #314
• feat(agents): HTTP / remote agent client (6/8) by @pavelgj in #315
• test(agents): cross-SDK conformance suite (7/8) by @pavelgj in #316
• docs(agents): demo server + web UI (8/8) by @pavelgj in #317
• feat(google-cloud): add genkit_google_cloud with a FirestoreSessionStore by @pavelgj in #318
• feat(middleware): add agents sub-agent delegation middleware by @pavelgj in #324
• feat(genkit): type-safe agent State with schemantic parsing + agent API polish by @pavelgj in #330
• docs(testapps/agents): add reasoning display and file viewer to agents app by @pavelgj in #334

Other changes

Providers & models

• feat(genkit_firebase_ai): Support Vertex AI Gemini API by @goderbauer in #296
• feat(genkit_vertexai): support Gemini and multimodal embedders by @CorieW in #261
• feat(vertexai): curated known-model metadata + P0 Gemini 3.x registrations by @cabljac in #320
• refactor(google_genai,vertexai): model curated Gemini catalog as an enum by @cabljac in #323
• fix(genkit_openai): send non-image media as OpenAI file content parts by @irangarcia in #297

Core & prompts

• feat!: pass middleware context with GenkitAI to factories by @pavelgj in #319
• fix(core): forward init (and context) from reflection runAction to actions by @pavelgj in #321
• fix(prompt): carry resolved middleware refs on rendered options by @pavelgj in #325
• feat(prompt): surface middleware and generate options in prompt loader by @pavelgj in #333

Schemantic

• feat(schemantic): add serialize to convert typed values to JSON by @pavelgj in #332
• docs: add documentation to schemantic_builder entry point by @pavelgj in #294

Breaking changes

• feat!: pass middleware context with GenkitAI to factories (#319) - middleware factories now receive the middleware context via GenkitAI.

New Contributors

• @goderbauer made their first contribution in #296

Full Changelog: genkit-v0.14.1...genkit-v0.15.0

Breaking Changes

• pass middleware context with GenkitAI to factories (#319)

Features

• surface middleware and generate options in prompt loader (#333)
• type-safe agent State with schemantic parsing + agent API polish (#330)
• HTTP / remote agent client (6/8) (#315)
• server-side agent runtime (5/8) (#314)
• transport-agnostic client core (4/8) (#313)
• dart:io FileSessionStore (3/8) (#312)
• session & snapshot storage + JSON Patch (2/8) (#311)
• agent & session schema types (1/8) (#310)

Fixes

• carry resolved middleware refs on rendered options (#325)
• forward init (and context) from reflection runAction to actions (#321)

Other Changes

• cross-SDK conformance suite (7/8) (#316)

Fixes

• apply input and output schemas when loading .prompt files (#300)

Breaking Changes

• add dotprompt integration with executable prompts and folder loading (#279)

Features

• add top-level system parameter to generate() (#289)
• Add interrupt releated span metadata (#287)

Other Changes

• update model references to gemini-flash-latest (#293)
• split schemantic into runtime and schemantic_builder packages (#292)

Features

• add listValues (#281)

Fixes

• preserve middleware in generation (#282)

Features

• generate docs for generated schemantic types (#274)

Breaking Changes

• introduce GenerateTurnState to middleware generate hook and improve chunk indexing (#269)

Features

• add additionalProperties support to @Schema and implement strict object validation (#251)

Features

• support middleware and defaultModel in listValues (#240)

Fixes

• updated reflection v2 implementation in line with latest spec changes (#231)

Breaking Changes

• changed middleware tool hook return type to Part for greater flexibility (#218)

Features

• Cache plugin action lists using a new adapter to optimize discovery (#217)

Fixes

• correctly handle enums in the generated constructor (#220)

Other Changes

• minor refactor to avoid use of dynamic for tool status (#219)

Features

• Allow asynchronous header generation for remote models (#212)

Breaking Changes

• changed tool hook signature on middleware, pass toolRequest to tool (#211)

• FEAT: Add support for a default model and make generate model param optional (#203).
• FEAT: added remoteModel for defining remote AI models with lite api (#198).
• DOCS: update docs, example, package descriptions and regen types (#201).

2026-03-05

Changes

Packages with breaking changes:

• genkit_google_genai - v0.2.0
• genkit_vertexai - v0.2.0

Packages with other changes:

• There are no other changes in this release.

• FEAT: Add support for a default model and make generate model param optional (#203).
• FEAT: added remoteModel for defining remote AI models with lite api (#198).
• DOCS: update docs, example, package descriptions and regen types (#201).

• Graduate package to a stable release. See pre-releases prior to this version for changelog entries.

• FEAT: added generate span (#196).
• FEAT: Enhance extract function to support primitive JSON types (string and numbers) (#195).

2026-03-03

Changes

Packages with breaking changes:

• There are no breaking changes in this release.

Packages with other changes:

• genkit_google_genai - v0.0.1-dev.19

• FEAT: added generate span (#196).
• FEAT: Enhance extract function to support primitive JSON types (string and numbers) (#195).

• REFACTOR: Tweak RegExps and avoid non-linear complexity (#175).
• REFACTOR: make all classes final or base (#179).
• REFACTOR: centralize status-to-http mapping for transport errors (#181).
• FEAT: introduce Genkit evaluation functionality (#191).
• FEAT(openai): add Vertex support with shared Vertex auth utilities (#185).
• BREAKING REFACTOR: renamed @Schematic() to @Schema() (#192).
• BREAKING FEAT: introduced dynamic action provider and migrated MCP plugin to use DAP (#187).

Note: This release has breaking changes.

• REFACTOR: Tweak RegExps and avoid non-linear complexity (#175).
• REFACTOR: make all classes final or base (#179).
• REFACTOR: centralize status-to-http mapping for transport errors (#181).
• FEAT: introduce Genkit evaluation functionality (#191).
• FEAT(openai): add Vertex support with shared Vertex auth utilities (#185).
• BREAKING REFACTOR: renamed @Schematic() to @Schema() (#192).
• BREAKING FEAT: introduced dynamic action provider and migrated MCP plugin to use DAP (#187).

• REFACTOR: hide package:json_schema_builder (#167).
• FIX: do not default instructions for json format (should use native constrained generation) (#176).
• FIX: enable and fix a couple of lints (#174).
• FIX: enforce formatting check in CI (#166).
• FIX: be consistent with String quotes (#164).
• FIX: fix strict casts (#165).
• FIX: don't import dart:io in registry (#159).
• FIX: lite.dart needs to call the function (#160).
• FIX: better generics (#153).
• FIX: move Genkit class to a library and export (#152).
• FIX: fix a couple of dartdoc issues (#151).
• FIX: extractJson return null for partial mode when no JSON started (#141).
• BREAKING FEAT: move basic type functions to static creation method on SchemanticType (#154).

Note: This release has breaking changes.

• REFACTOR: hide package:json_schema_builder (#167).
• FIX: do not default instructions for json format (should use native constrained generation) (#176).
• FIX: enable and fix a couple of lints (#174).
• FIX: enforce formatting check in CI (#166).
• FIX: be consistent with String quotes (#164).
• FIX: fix strict casts (#165).
• FIX: don't import dart:io in registry (#159).
• FIX: lite.dart needs to call the function (#160).
• FIX: better generics (#153).
• FIX: move Genkit class to a library and export (#152).
• FIX: fix a couple of dartdoc issues (#151).
• FIX: extractJson return null for partial mode when no JSON started (#141).
• BREAKING FEAT: move basic type functions to static creation method on SchemanticType (#154).

• REFACTOR: Introduce a dedicated plugin.dart entry point for plugin-related exports (#149).
• FEAT: improve partial json extraction (#150).
• FEAT: add error handling for plugin action listing and report failures to stderr (#148).
• FEAT: Allow ReflectionServerV1 to automatically find an available port if none is specified. (#146).

• FIX: prevent incorrect partial JSON repair by validating stack state (#144).
• FEAT: Add remote model support and enable serving actions via shelf (#143).

• REFACTOR: automate telemetry exporter configuration (#131).
• FEAT: implemented/fixed tools calling and structured output for firebase_ai (#138).

• FIX: Wrap error responses in a JSON object under an 'error' key (#130).
• FEAT: Implemented real-time tracing (#128).
• FEAT: created a genkit_middleware package with skills, filesystem and toolApproval middleware (#126).
• FEAT: add MCP (Model Context Protocol) plugin (#94).
• FEAT: implemented interrupt restart (#124).
• BREAKING REFACTOR: generate api cleanup (#125).

Note: This release has breaking changes.

• FIX: Wrap error responses in a JSON object under an 'error' key (#130).
• FEAT: Implemented real-time tracing (#128).
• FEAT: created a genkit_middleware package with skills, filesystem and toolApproval middleware (#126).
• FEAT: add MCP (Model Context Protocol) plugin (#94).
• FEAT: implemented interrupt restart (#124).
• BREAKING REFACTOR: generate api cleanup (#125).

• FEAT: introducing registered middleware (#87).
• FEAT: added support for embedders (embedding models) (#88).

• FIX: Coerce num values to double for generated double fields during JSON parsing. (#65).
• FEAT: add Google Search and multi-speaker voice config support, extract usage metadata, and introduce reasoning parts (#82).
• FEAT: allow generate and generateBidi to accept Tool objects directly in the tools list alongside tool names (#79).
• FEAT: Implement hierarchical registry with parent delegation and merging for values and actions (#78).
• FEAT: Implement streaming chunk indexing across turns and improve maxTurns error handling with a new default. (#75).
• FEAT: implemented interrupts (#73).
• FEAT: Add retry middleware for AI model and tool calls with configurable backoff and error handling. (#67).
• FEAT: Add $GenerateResponse type, refine schema types, and update generated class constructors to use late final and regular constructors. (#66).
• FEAT: added schemas for gemini models, made sure TTS and nano banana models are working (#63).
• BREAKING REFACTOR: update GenkitException to use a StatusCodes enum instead of raw integer status codes. (#68).

Note: This release has breaking changes.

• FIX: Coerce num values to double for generated double fields during JSON parsing. (#65).
• FEAT: add Google Search and multi-speaker voice config support, extract usage metadata, and introduce reasoning parts (#82).
• FEAT: allow generate and generateBidi to accept Tool objects directly in the tools list alongside tool names (#79).
• FEAT: Implement hierarchical registry with parent delegation and merging for values and actions (#78).
• FEAT: Implement streaming chunk indexing across turns and improve maxTurns error handling with a new default. (#75).
• FEAT: implemented interrupts (#73).
• FEAT: Add retry middleware for AI model and tool calls with configurable backoff and error handling. (#67).
• FEAT: Add $GenerateResponse type, refine schema types, and update generated class constructors to use late final and regular constructors. (#66).
• FEAT: added schemas for gemini models, made sure TTS and nano banana models are working (#63).
• BREAKING REFACTOR: update GenkitException to use a StatusCodes enum instead of raw integer status codes. (#68).

• FEAT: updated AnyOf support for union types in Schemantic, including helper class generation and schema type handling. (#62).

• REFACTOR: reimplement schema generation from extension types to classes, enhance PartExtension getters, and simplify GenerateResponse and tool invocation. (#53).
• FEAT: use combining builder and header option (#52).
• BREAKING FEAT: implement Schemantic API redesign with $ prefixed schema definitions and static $schema for unified schema access. (#60).

Note: This release has breaking changes.

• REFACTOR: reimplement schema generation from extension types to classes, enhance PartExtension getters, and simplify GenerateResponse and tool invocation. (#53).
• FEAT: use combining builder and header option (#52).
• BREAKING FEAT: implement Schemantic API redesign with $ prefixed schema definitions and static $schema for unified schema access. (#60).

• BREAKING REFACTOR: renamed JsonExtensionType to SchemanticType (#44).

Note: This release has breaking changes.

• BREAKING REFACTOR: renamed JsonExtensionType to SchemanticType (#44).

• REFACTOR: Consolidate Google GenAI examples into a single file, fixed tools calling, and schema flattening helper (#43).
• FEAT: implemented streaming and various config options for genkit_google_genai plugin (#42).

• BREAKING FEAT: Refactor basic types into factory functions to support schema constraints (#34).

Note: This release has breaking changes.

• BREAKING FEAT: Refactor basic types into factory functions to support schema constraints (#34).

• REFACTOR: move the package-specific schema generator into a peer package (#31).
• BREAKING REFACTOR: renamed @Key annotation to @Field (#30).

Note: This release has breaking changes.

• REFACTOR: move the package-specific schema generator into a peer package (#31).
• BREAKING REFACTOR: renamed @Key annotation to @Field (#30).

• REFACTOR: make generated JsonExtensionType factory classes (*TypeFactory) private (#29).
• FEAT: added support for defining listType and mapType in schemantic (#28).

• FEAT: bump analyzer dependency (#25).
• FEAT: added support for schema refs/defs in the schema generator (#22).
• BREAKING REFACTOR: renamed genkit_schema_builder package to schemantic (#26).

Note: This release has breaking changes.

• FEAT: bump analyzer dependency (#25).
• FEAT: added support for schema refs/defs in the schema generator (#22).
• BREAKING REFACTOR: renamed genkit_schema_builder package to schemantic (#26).

• FIX: register generate action with the correct name.
• FEAT: implemented live api using firebase ai logic (#19).

• Initial release of Genkit Dart framework.
• BREAKING CHANGE: RemoteAction has 2 extra generic type parameters I and Init for the input and init types.
• feat: defineRemoteAction now accepts inputType, outputType and streamType parameters using genkit schema builder types.

• Made fromResponse and fromStreamChunk optional in defineRemoteAction. If not provided, the response and stream chunks will be dynamic objects decoded from JSON, instead of requiring a typed conversion function.

• BREAKING CHANGE: The .stream() method now returns an ActionStream instead of a FlowStreamResponse record. ActionStream is a Stream that provides two ways to access the flow's final, non-streamed response:

  • onResult: A Future that completes with the result. This is the recommended approach. It will complete with a GenkitException if the stream terminates with an error or is cancelled.
  • result: A synchronous getter that should only be used after the stream is fully consumed. It will throw a GenkitException if the stream is not consumed, terminates with an error or is cancelled.

  Migration: Code that previously looked like this:

  final (:stream, :response) = myAction.stream(input: ...);
  await for (final chunk in stream) {
    // ...
  }
  final finalResult = await response;
  

  Should be updated to use onResult:

  final stream = myAction.stream(input: ...);
  await for (final chunk in stream) {
    // ...
  }
  final finalResult = await stream.onResult;
  // or
  final finalResult = stream.result;
  

• BREAKING CHANGE: The package has been renamed from package:genkit/genkit.dart to package:genkit/client.dart. You will need to update your import statements.
• BREAKING CHANGE: The response future returned by the .stream() method is now nullable (Future<O?>). This change supports improved error handling and cancellation.
• Improved Error Handling: Errors occurring on the server during a stream are now thrown by the stream itself. This allows you to catch exceptions directly within a try/catch block surrounding an await for loop.

• Added standard Genkit data classes for working with generative models, including GenerateResponse, Message, and Part types.
• Added helper getters like .text and .media for easier data extraction.

• README cleanup

• Enhanced type-safe client: Added comprehensive generics support for type-safe operations
• Streaming support: Implemented real-time data streaming with Server-Sent Events (SSE)
• Improved error handling: Introduced GenkitException with detailed error information and HTTP status codes
• Authentication support: Added support for custom headers including Firebase Auth integration
• Better integration: Enhanced compatibility with json_serializable for object serialization
• Comprehensive documentation: Added detailed API documentation and usage examples
• Platform support: Full cross-platform support for iOS, Android, Web, Windows, macOS, and Linux
• Testing improvements: Added comprehensive unit and integration tests

• Initial version.