ai_sdk_remote

Optional Dart transport for trusted backends that emit the Vercel AI SDK 7.0.111 UI-message stream protocol. It converts conversation history to the validated UIMessage request shape, consumes version v1 SSE responses, and reduces text, reasoning, tool, approval, source, file, metadata, error, and unknown events into immutable ai_sdk_conversation snapshots.

final transport = RemoteConversationTransport(
  endpoint: Uri.parse('https://example.test/api/chat'),
  authHeaders: () => {'Authorization': 'Bearer $sessionToken'},
);
final snapshots = transport.send(conversation);

The package does not execute tools, contain provider keys, retry, replay, or claim stream resumption. A response must use text/event-stream, the x-vercel-ai-ui-message-stream: v1 header, include finish or abort, and terminate with data: [DONE]. Cancellation uses http.AbortableRequest and disposal aborts active requests. Applications own persistence and server-side authorization; injected HTTP clients remain owned by the caller.

The transport requires http 1.6 or newer for abortable requests and the browser fix for cancelling a response while waiting for its next chunk.

Run the included Dart example against the repository's pinned reference server:

dart run example/example.dart http://127.0.0.1:8081/chat

The reference server lives in examples/remote_backend/js at the repository root. Run npm ci and node server.mjs there first. It uses static events and requires no provider key. The Dart example sends a typed conversation, prints updated assistant text, and disposes its owned transport when the stream ends.