StreamProcessor

Defined in: packages/ai/src/activities/chat/stream/processor.ts:237

StreamProcessor - State machine for processing AI response streams

Manages the full UIMessage[] conversation and emits events on changes. Trusts the adapter contract: adapters emit clean AG-UI events in the correct order.

State tracking:

• Full message array
• Per-message stream state (text, tool calls, thinking)
• Multiple concurrent message streams
• Tool call completion via TOOL_CALL_END events

See

• docs/chat-architecture.md#streamprocessor-internal-state — State field reference
• docs/chat-architecture.md#adapter-contract — What this class expects from adapters

Constructors

Constructor

new StreamProcessor(options?): StreamProcessor;

new StreamProcessor(options?): StreamProcessor;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:307

Parameters

options?

StreamProcessorOptions = {}

Returns

StreamProcessor

Methods

addToolApprovalResponse()

addToolApprovalResponse(approvalId, approved): void;

addToolApprovalResponse(approvalId, approved): void;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:511

Add an approval response (called by client after handling onApprovalRequest)

Parameters

approvalId

string

approved

boolean

Returns

void

addToolResult()

addToolResult(
   toolCallId, 
   output, 
   error?): void;

addToolResult(
   toolCallId, 
   output, 
   error?): void;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:460

Add a tool result (called by client after handling onToolCall)

Parameters

toolCallId

string

output

any

error?

string

Returns

void

addUserMessage()

addUserMessage(
   content, 
   id?, 
   metadata?): UIMessage;

addUserMessage(
   content, 
   id?, 
   metadata?): UIMessage;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:387

Add a user message to the conversation. Supports both simple string content and multimodal content arrays.

Parameters

content

string | ContentPart[]

The message content (string or array of content parts)

id?

string

Optional custom message ID (generated if not provided)

metadata?

Record<string, any>

Optional AG-UI metadata bag

Returns

UIMessage

The created UIMessage

Example

// Simple text message
processor.addUserMessage('Hello!')

// Multimodal message with image
processor.addUserMessage([
  { type: 'text', content: 'What is in this image?' },
  { type: 'image', source: { type: 'url', value: 'https://example.com/photo.jpg' } }
])

// With custom ID
processor.addUserMessage('Hello!', 'custom-id-123')

// Simple text message
processor.addUserMessage('Hello!')

// Multimodal message with image
processor.addUserMessage([
  { type: 'text', content: 'What is in this image?' },
  { type: 'image', source: { type: 'url', value: 'https://example.com/photo.jpg' } }
])

// With custom ID
processor.addUserMessage('Hello!', 'custom-id-123')

areAllToolsComplete()

areAllToolsComplete(): boolean;

areAllToolsComplete(): boolean;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:551

Check if all tool calls in the last assistant message are complete Useful for auto-continue logic

Returns

boolean

clearMessages()

clearMessages(): void;

clearMessages(): void;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:642

Clear all messages

Returns

void

finalizeStream()

finalizeStream(): void;

finalizeStream(): void;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:3642

Finalize the stream — complete all pending operations.

Called when the async iterable ends (stream closed). Acts as the final safety net: completes any remaining tool calls, flushes un-emitted text, and fires onStreamEnd.

Returns

void

See

docs/chat-architecture.md#single-shot-text-response — Finalization step

getCurrentAssistantMessageId()

getCurrentAssistantMessageId(): string | null;

getCurrentAssistantMessageId(): string | null;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:444

Get the current assistant message ID (if one has been created). Returns null if prepareAssistantMessage() was called but no content has arrived yet.

Returns

string | null

getMessages()

getMessages(): UIMessage<unknown>[];

getMessages(): UIMessage<unknown>[];

Defined in: packages/ai/src/activities/chat/stream/processor.ts:543

Get current messages

Returns

UIMessage<unknown>[]

getRecording()

getRecording(): ChunkRecording | null;

getRecording(): ChunkRecording | null;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:3810

Get the current recording

Returns

ChunkRecording | null

getState()

getState(): ProcessorState;

getState(): ProcessorState;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:3767

Get current processor state (aggregated across all messages)

Returns

ProcessorState

prepareAssistantMessage()

prepareAssistantMessage(): void;

prepareAssistantMessage(): void;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:422

Prepare for a new assistant message stream. Does NOT create the message immediately -- the message is created lazily when the first content-bearing chunk arrives via ensureAssistantMessage(). This prevents empty assistant messages from flickering in the UI when auto-continuation produces no content.

Returns

void

prependMessages()

prependMessages(messages): void;

prependMessages(messages): void;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:348

Put older UI messages at the front of the conversation.

Skip a message if its id is already in the list. Keep the existing message. Then emit the same messages-change event as setMessages.

Use this for older history pages. The first hydrate window uses setMessages.

Parameters

messages

UIMessage<unknown>[]

Older UI messages in insertion order. The first item is the oldest.

Returns

void

Example

processor.setMessages([newest])
processor.prependMessages([oldest])

processor.setMessages([newest])
processor.prependMessages([oldest])

process()

process(stream): Promise<ProcessorResult>;

process(stream): Promise<ProcessorResult>;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:669

Process a stream and emit events through handlers

Parameters

stream

AsyncIterable<any>

Returns

Promise<ProcessorResult>

processChunk()

processChunk(chunk): void;

processChunk(chunk): void;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:703

Process a single chunk from the stream.

Central dispatch for all AG-UI events. Each event type maps to a specific handler. Events not listed in the switch are intentionally ignored (STATE_SNAPSHOT, STATE_DELTA).

Parameters

chunk

AGUIEvent

Returns

void

See

docs/chat-architecture.md#adapter-contract — Expected event types and ordering

removeMessagesAfter()

removeMessagesAfter(index): void;

removeMessagesAfter(index): void;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:591

Remove messages after a certain index (for reload/retry)

Parameters

index

number

Returns

void

reset()

reset(): void;

reset(): void;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:3842

Full reset (including messages)

Returns

void

setMessages()

setMessages(messages): void;

setMessages(messages): void;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:327

Set the messages array (e.g., from persisted state)

Parameters

messages

UIMessage<unknown>[]

Returns

void

startAssistantMessage()

startAssistantMessage(messageId?): string;

startAssistantMessage(messageId?): string;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:431

Parameters

messageId?

string

Returns

string

Deprecated

Use prepareAssistantMessage() instead. This eagerly creates an assistant message which can cause empty message flicker.

startRecording()

startRecording(): void;

startRecording(): void;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:3797

Start recording chunks

Returns

void

toModelMessages()

toModelMessages(): ModelMessage<
  | string
  | ContentPart<unknown, unknown, unknown, unknown, unknown>[]
  | null>[];

toModelMessages(): ModelMessage<
  | string
  | ContentPart<unknown, unknown, unknown, unknown, unknown>[]
  | null>[];

Defined in: packages/ai/src/activities/chat/stream/processor.ts:532

Get the conversation as ModelMessages (for sending to LLM)

Returns

ModelMessage< | string | ContentPart<unknown, unknown, unknown, unknown, unknown>[] | null>[]

replay()

static replay(recording, options?): Promise<ProcessorResult>;

static replay(recording, options?): Promise<ProcessorResult>;

Defined in: packages/ai/src/activities/chat/stream/processor.ts:3863

Replay a recording through the processor

Parameters

recording

ChunkRecording

options?

StreamProcessorOptions

Returns

Promise<ProcessorResult>