A better way to handle Agent stream output for client display

[KanchiShimono]

Context

I want to return the Agent’s (LLM’s) stream output to the client token by token for users of my chat UI.
From what I understand, the most straightforward way to achieve this using the A2A protocol is to include the tokens in TaskStatusUpdateEvent.status.message in response to a message/stream request.

However, the current Python and TypeScript SDK implementations have the following issues:

1. Task.history records the history token by token, making it cumbersome to reconstruct conversation history across multiple turns.
2. Each token output triggers access to TaskStore, which leads to heavy write load when using a database-backed TaskStore.

I believe it’s preferable to store conversation history as a single message per complete Agent response.

Question

1. Given this, is there a good way to display the Agent’s token output on the client?
2. Has there been any consideration of introducing a new type dedicated to display (e.g., something like MessageDelta) that does not update task state?

[acensia]

I'm having the same question.
I've tried to send/show each generation of tokens (or chunks) with each task or artifact, but both seems not plausible.