Class TubeBackedStreamingChatResponseHandler

java.lang.Object

dev.langchain4j.reactive.streaming.TubeBackedStreamingChatResponseHandler

All Implemented Interfaces:

StreamingChatResponseHandler

public final class TubeBackedStreamingChatResponseHandler
extends Object
implements StreamingChatResponseHandler

Bridge from the handler-based StreamingChatResponseHandler contract to a Tube of ChatModelStreamingEvents. It lets the very same dispatch logic that drives the handler-based streaming API drive the reactive publisher-based API, with no duplicated event-mapping code. Shared by all streaming chat model providers.

Each handler callback maps to a tube.send(event); onCompleteResponse also terminates the tube; onError fails it. Calls after cancellation are silently dropped.

Cancellation. Two provider wirings are supported:

• Subscription-based — providers streaming over HttpClient.stream() (e.g. OpenAI, Anthropic): the reactive subscriber cancels its Flow.Subscription from tube.whenTerminates(...). This bridge only supplies a non-null streamingHandle() for dispatchers that must put one in their per-callback contexts; its cancel() is a no-op (cancellation goes through the subscription).
• Handle-based — providers whose only abort mechanism is a StreamingHandle threaded through the per-callback contexts (e.g. Bedrock over the AWS SDK): the bridge captures that handle lazily and cancels it via cancelUpstream(), which the provider wires to tube.whenTerminates(...).

Constructor Summary

Constructors

Constructor	Description
TubeBackedStreamingChatResponseHandler(mutiny.zero.Tube<ChatModelStreamingEvent> tube)

Method Summary

Modifier and Type	Method	Description
void	cancelUpstream()	Aborts the upstream stream if a genuine StreamingHandle was captured from the per-callback contexts (handle-based providers).
void	onCompleteResponse(ChatResponse completeResponse)	Invoked when the model has finished streaming a response.
void	onCompleteToolCall(CompleteToolCall completeToolCall)	Invoked when the model has finished streaming a single tool call.
void	onError(Throwable error)	This method is invoked when an error occurs during streaming.
void	onPartialResponse(PartialResponse partialResponse, PartialResponseContext context)	Invoked each time the model generates a partial textual response, usually a single token.
void	onPartialThinking(PartialThinking partialThinking, PartialThinkingContext context)	Invoked each time the model generates a partial thinking/reasoning text, usually a single token.
void	onPartialToolCall(PartialToolCall partialToolCall, PartialToolCallContext context)	This callback is invoked each time the model generates a partial tool call, which contains a single token of the tool's arguments.
void	onUnmappedRawEvent(Object rawEvent)	Invoked when a provider emits a raw streaming event that is not already exposed through one of the typed callbacks (such as StreamingChatResponseHandler.onPartialResponse(PartialResponse, PartialResponseContext), StreamingChatResponseHandler.onPartialThinking(PartialThinking, PartialThinkingContext), StreamingChatResponseHandler.onPartialToolCall(PartialToolCall, PartialToolCallContext) or StreamingChatResponseHandler.onCompleteToolCall(CompleteToolCall)).
StreamingHandle	streamingHandle()	A non-null StreamingHandle for dispatchers that must supply one in their per-callback contexts.

Methods inherited from class Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Methods inherited from interface StreamingChatResponseHandler

onPartialResponse, onPartialThinking, onPartialToolCall

Constructor Details

TubeBackedStreamingChatResponseHandler

public TubeBackedStreamingChatResponseHandler(mutiny.zero.Tube<ChatModelStreamingEvent> tube)

Method Details

streamingHandle

public StreamingHandle streamingHandle()

A non-null StreamingHandle for dispatchers that must supply one in their per-callback contexts. Its cancel() is a no-op; subscription-based providers cancel via Flow.Subscription instead.

cancelUpstream

public void cancelUpstream()

Aborts the upstream stream if a genuine StreamingHandle was captured from the per-callback contexts (handle-based providers). Wire this to tube.whenTerminates(...) so a downstream cancellation (or a buffer overflow) tears down the upstream instead of draining it. Idempotent; a no-op for subscription-based providers, which never supply an upstream handle here.

onPartialResponse

public void onPartialResponse(PartialResponse partialResponse,
 PartialResponseContext context)

Description copied from interface: StreamingChatResponseHandler

Invoked each time the model generates a partial textual response, usually a single token.

Please note that some LLM providers do not stream individual tokens, but send responses in batches. In such cases, this callback may receive multiple tokens at once.

Either this or the StreamingChatResponseHandler.onPartialResponse(String) method should be implemented if you want to consume tokens as soon as they become available.

Specified by:

onPartialResponse in interface StreamingChatResponseHandler

Parameters:

partialResponse - A partial textual response, usually a single token.

context - A partial response context. Contains a StreamingHandle that can be used to cancel streaming.

See Also:

• StreamingChatResponseHandler.onPartialResponse(String)

onPartialThinking

public void onPartialThinking(PartialThinking partialThinking,
 PartialThinkingContext context)

Description copied from interface: StreamingChatResponseHandler

Invoked each time the model generates a partial thinking/reasoning text, usually a single token.

Please note that some LLM providers do not stream individual tokens, but send thinking tokens in batches. In such cases, this callback may receive multiple tokens at once.

Either this or the StreamingChatResponseHandler.onPartialThinking(PartialThinking) method should be implemented if you want to consume thinking tokens as soon as they become available.

Specified by:

onPartialThinking in interface StreamingChatResponseHandler

Parameters:

partialThinking - A partial thinking text, usually a single token.

context - A partial thinking context. Contains a StreamingHandle that can be used to cancel streaming.

See Also:

• StreamingChatResponseHandler.onPartialThinking(PartialThinking)

onPartialToolCall

public void onPartialToolCall(PartialToolCall partialToolCall,
 PartialToolCallContext context)

Description copied from interface: StreamingChatResponseHandler

This callback is invoked each time the model generates a partial tool call, which contains a single token of the tool's arguments. It is typically invoked multiple times for a single tool call until StreamingChatResponseHandler.onCompleteToolCall(CompleteToolCall) is eventually invoked, indicating that the streaming for that tool call is finished.

Here's an example of what streaming a single tool call might look like:

1. onPartialToolCall(index = 0, id = "call_abc", name = "get_weather", partialArguments = "{\"")
2. onPartialToolCall(index = 0, id = "call_abc", name = "get_weather", partialArguments = "city")
3. onPartialToolCall(index = 0, id = "call_abc", name = "get_weather", partialArguments = ""\":\"")
4. onPartialToolCall(index = 0, id = "call_abc", name = "get_weather", partialArguments = "Mun")
5. onPartialToolCall(index = 0, id = "call_abc", name = "get_weather", partialArguments = "ich")
6. onPartialToolCall(index = 0, id = "call_abc", name = "get_weather", partialArguments = "\"}")
7. onCompleteToolCall(index = 0, id = "call_abc", name = "get_weather", arguments = "{\"city\":\"Munich\"}")

If the model decides to call multiple tools, the index will increment, allowing you to correlate.

Please note that not all LLM providers stream tool calls token by token. Some providers (e.g., Bedrock, Google, Mistral, Ollama) return only complete tool calls. In those cases, this callback won't be invoked - only StreamingChatResponseHandler.onCompleteToolCall(CompleteToolCall) will be called.

Either this or the StreamingChatResponseHandler.onPartialToolCall(PartialToolCall) method should be implemented if you want to consume partial tool calls as soon as they become available.

Specified by:

onPartialToolCall in interface StreamingChatResponseHandler

Parameters:

partialToolCall - A partial tool call that contains the index, tool ID, tool name and partial arguments.

context - A partial tool call context. Contains a StreamingHandle that can be used to cancel streaming.

See Also:

• StreamingChatResponseHandler.onPartialToolCall(PartialToolCall)

onCompleteToolCall

public void onCompleteToolCall(CompleteToolCall completeToolCall)

Description copied from interface: StreamingChatResponseHandler

Invoked when the model has finished streaming a single tool call.

Specified by:

onCompleteToolCall in interface StreamingChatResponseHandler

Parameters:

completeToolCall - A complete tool call that contains the index, tool ID, tool name, and fully assembled arguments.

onUnmappedRawEvent

public void onUnmappedRawEvent(Object rawEvent)

Description copied from interface: StreamingChatResponseHandler

Invoked when a provider emits a raw streaming event that is not already exposed through one of the typed callbacks (such as StreamingChatResponseHandler.onPartialResponse(PartialResponse, PartialResponseContext), StreamingChatResponseHandler.onPartialThinking(PartialThinking, PartialThinkingContext), StreamingChatResponseHandler.onPartialToolCall(PartialToolCall, PartialToolCallContext) or StreamingChatResponseHandler.onCompleteToolCall(CompleteToolCall)).

This acts as an escape hatch for provider-specific events that langchain4j does not model, such as server-tool lifecycle events (e.g., OpenAI's web_search_call.in_progress). Events that are already delivered as partial responses, thinking or tool calls are not repeated here.

The event type depends on the provider implementation. Implementations using the dev.langchain4j.http.client.HttpClient abstraction (e.g., OpenAI, Anthropic, Google AI Gemini) typically expose ServerSentEvent; other implementations can expose provider-specific event objects (e.g., the OpenAI official Responses model exposes the SDK's ResponseStreamEvent).

Specified by:

onUnmappedRawEvent in interface StreamingChatResponseHandler

Parameters:

rawEvent - A raw provider streaming event.

onCompleteResponse

public void onCompleteResponse(ChatResponse completeResponse)

Description copied from interface: StreamingChatResponseHandler

Invoked when the model has finished streaming a response.

Specified by:

onCompleteResponse in interface StreamingChatResponseHandler

Parameters:

completeResponse - The complete response generated by the model, containing all assembled partial text and tool calls.

onError

public void onError(Throwable error)

Description copied from interface: StreamingChatResponseHandler

This method is invoked when an error occurs during streaming.

Specified by:

onError in interface StreamingChatResponseHandler

Parameters:

error - The error that occurred