Releases
v1.15.0
Compare
Sorry, something went wrong.
No results found
1.15.0 (2026-11-18)
Features
add FunctionNode, a workflow step with typed Java ports (d627147 )
add TypeReference ports and a build-time port check to FunctionNode (01f899f )
check workflow edges by assignability (a0cc1dc )
classifiers: add ClassifierNode for workflow routing (61e0eb8 )
classifiers: add LlmClassifier backed by BaseLlm (f3e1cea )
classifiers: add the experimental Classifier interface and its question and answer types (739c231 )
classifiers: read and write questions and answers as JSON (ef5f798 )
dev: serve workflow graph images (677557b )
hide another node's input from an agent node (0f1b2c8 )
infer schemas for FunctionNode ports (020c95c )
keep a SINGLE_TURN LlmAgent node to its own node path (f14e184 )
let models report whether they take an output schema with tools (5ce839b )
make a SINGLE_TURN LlmAgent node's answer its output (d122338 )
mcp: add opt-in CallToolResultMapper for tool results (4dfe0b5 )
models: bring Claude to parity with ADK Python (62003c1 )
run a SINGLE_TURN LlmAgent node on its node input (e264177 )
run a SINGLE_TURN LlmAgent node without history by default (8e56693 )
tokt: carry transcriptions and turnComplete across the interop (30edd99 )
tokt: convert speech and transcription configs (ab34617 )
tokt: run agents through the ADK Java runner's runLive (3582f1a )
tokt: send blobs and run a held session through runLive (1f1537d )
tokt: stream a Flowable FunctionTool in a live run (42b2f9c )
workflow: add JoinNode.create for a join that needs only a name (d100738 )
workflow: ask the user to approve a FunctionNode run (1b11290 )
workflow: check an untyped enum on an edge as the number or boolean it spells (beda447 )
workflow: interrupt a canceled FunctionNode's body (3c07372 )
workflow: let a FunctionNode body ask for input with requestInput (61ce022 )
workflow: let a FunctionNode body send events with ctx.emit (c90ac7c )
workflow: reject port types that ADK Kotlin rejects, such as ZoneId (8ddd5b4 )
workflow: run a FunctionNode body on the node's scheduler (d0c3338 )
Bug Fixes
a2a: continue the remote context on the next turn (5ade857 )
a2a: keep the metadata of a message reply (efedf4c )
a2a: read the error code ADK Python and Go agents send (659f5ab )
a2a: resume a remote agent's task when the user answers its call (7c80a14 )
agents: stop saving a transfer target's reply to the parent's outputKey (abb86a7 )
aggregate live text chunks and skip partial events in runLive (9435377 )
count events after the last model response toward the compaction threshold (587d252 )
declare set_model_response parameters as an object for untyped schemas (a33af4c )
dev: listen on 127.0.0.1 by default in AdkWebServer.main() and start() (464cf48 )
flows: run onModelErrorCallback when a model throws synchronously (e5cd285 )
flows: share the persist barrier with copies of callbackContextData (f9ec28c )
gemini: build lists from streamed function call argument paths (def4ffa )
keep apigee/gemini/ models on the Gemini API even when the environment selects Vertex AI (1f7efe3 )
keep whole numbers in streamed function call arguments as integers (be4984e )
langchain4j: accept free-form schemas in tool declarations (8c15ce5 )
langchain4j: end a streamed turn with one response that holds the whole answer (8a9b6d0 )
let a sub-agent transfer back to its parent in a resumed invocation (7fb0a3b )
live: answer a replayed history that ends in a user turn (f688c80 )
plugins: make BQAA GCS uploads create-only so transient errors retry (6a497da )
plugins: prevent BQAA offload data loss and preserve binary MIME (94b5bac )
show an agent the user's answer to another agent's call (658faba )
spring-ai: report cached prompt tokens in usage metadata (a4182a5 )
stop logging model input and output in Gemini (e10a7b6 )
tokt: forward model capabilities to the Kotlin engine (d3bd749 )
tokt: merge a Java plugin's replacement onto the original, as ADK Python does (0bb7f4a )
tools: keep the null of a required nullable property in set_model_response (e868293 )
use set_model_response on the Gemini API and for Vertex AI models that fail in JSON mode (faac2f4 )
validate free-form schemas instead of crashing (ad4ab50 )
workflow: don't start a canceled FunctionNode's confirmation predicate (c7f906c )
workflow: keep a node timeout from interrupting a FunctionNode's output (ea20742 )
workflow: keep a state write made while NodeRunner moves pending deltas (fca65ef )
workflow: keep the caller's trace context in a FunctionNode body (c5c238d )
workflow: pause a tool node when its tool asks for confirmation (f985064 )
workflow: run forked graph nodes concurrently when they block their threads (ba91fb8 )
workflow: run parallelNode items concurrently when they block their threads (ef1335e )
workflow: treat a null confirmed in a tool node's answer as a rejection (6e91348 )
workflow: wait for a canceled FunctionNode body before a retry (3fd6c1a )
wrap non-object output schemas for set_model_response (6973a79 )
Documentation
examples: add a classifier node workflow sample (87468e2 )
examples: add a workflows module with five samples that call no model (5217ce8 )
examples: add loop and nested workflow samples (47227c0 )
examples: add research and order-processing workflow samples (a968088 )
examples: add retry, returns and route workflow samples (a77e5dc )
examples: add samples that pass an agent's answer to the next node (84c0970 )
examples: add samples that run agents from a node and in parallel (59b4d1a )
examples: add workflow samples that pause for human input (51bc06c )
examples: run workflow-node agents in single-turn mode (7957d95 )
workflow: show how a FunctionNode body composes child nodes (f20d158 )
You can’t perform that action at this time.