gRPC YouTube disconnecting with EOF after a few seconds

I'm building a desktop application (Go + gRPC) to analyze YouTube Live Chat in real time. The REST endpoint liveChatMessages.list works, but I want to switch to liveChatMessages.streamList because according to the documentation it has lower quota usage.

I successfully established a gRPC connection and I receive chat messages normally, but after a few seconds the stream always closes with an EOF.

❓ Problem

The stream starts normally, receives a few responses, and then Recv() returns:

EOF

I don’t see any description in the documentation indicating that Google closes the stream periodically, or that clients must reconnect manually.

func (g *GrpcStreamListener) startStreaming(
    ctx context.Context,
    liveChatId string,
    onMessage func(*LiveChatMessage),
) error {

    streamCtx, cancel := context.WithCancel(ctx)
    g.cancel = cancel

    req := &LiveChatMessageListRequest{
        LiveChatId:       &liveChatId,
        Hl:               proto.String("pt-BR"),
        ProfileImageSize: proto.Uint32(64),
        MaxResults:       proto.Uint32(20),
        PageToken:        g.nextPageToken,
        Part: []string{"id", "snippet", "authorDetails"},
    }

    stream, err := g.client.StreamList(streamCtx, req)
    if err != nil {
        return fmt.Errorf("failed to start stream: %w", err)
    }

    g.stream = stream
    go g.receiveMessages(onMessage)
    return nil
}

func (g *GrpcStreamListener) receiveMessages(onMessage func(*LiveChatMessage)) {
    for {
        response, err := g.stream.Recv()
        if err != nil {
            // After a few seconds, err = EOF
            log.Printf("stream error: %v", err)
            break
        }

        g.nextPageToken = response.NextPageToken

        for _, msg := range response.GetItems() {
            onMessage(msg)
        }
    }
}

Google closes the stream by sending EOF after 5–10 seconds.

❓ Question

Is this the expected behavior? Does YouTube intentionally close the streamList connection after a short period and require clients to reconnect? Or is my implementation missing something?

If anyone has a working implementation or has already dealt with this EOF issue, any guidance is appreciated.