Streaming

Realtime callback

streamChatRealtime() calls your callback for each chunk as the provider sends it:

Generator

streamChat() returns a generator of chunks:

Some providers (such as OpenAI) collect the full response before yielding from streamChat(). Use streamChatRealtime() when you need output as it arrives.

The stream endpoint

POST /laravilt-ai/stream accepts:

Field	Type	Description
messages	array	role (system, user, assistant, tool) and content
provider	string	Optional provider name
model	string	Optional model
session_id	string	Optional session to save the exchange to
mentioned_resources	array	Optional resource slugs to focus on

It responds with text/event-stream. Each event is a JSON object with content (or error), and the stream ends with data: [DONE]:

Before streaming, the controller lets the model call the resource tools (list_resources, query_resource) and then streams the final answer.

Consuming the stream

The endpoint is a POST, so read it with fetch rather than EventSource:

The useAI() composable (Vue and React) wraps this for you; see Frontend components.