Clarification on missing name field in Responses API and handling multi-persona / multi-user dialogues

API
1

Hello OpenAI team,

I have a few questions regarding the recent Responses API schema.

In the former Chat Completions API, the message object supported a name field both for assistant and user roles.
However, in the new Responses API (input → input_message), the name field has been completely removed.

  1. Could you please clarify the reason for removing the name field?
    It was useful for distinguishing between multiple assistant personas (e.g. Planner, Executor, Critic) or multiple users within the same conversation.

  2. Do you plan to bring back name support in future versions of the Responses API?
    If so, is there a timeline or an alternative design planned?

  3. What is the recommended approach for handling multi-agent or multi-user flows now, using the Responses API?
    Specifically, cases where a single thread or context may include:

    • multiple assistant roles (Planner / Executor / Reviewer), or

    • messages coming from several users (User1, User2, User3) within the same conversation.

Any official guidance or examples on how you recommend structuring such input in the new API would be greatly appreciated.

Thank you very much for your time and clarification.

Best regards,
Vadim Ignatiev

2

The name parameter alongside role is useful for designating an input as an example, documentation, or adding the speaker’s name: 64 characters with alphanumeric and just hyphens or underscores.

Even yesterday, I had an example of how input could be improved by using the name field, as a substitute for roles that should exist with different hierarchy of natural understanding.

“name” could never be used for assigning who the assistant is, because the final “prompt” of assistant, automatically placed for the AI to complete its response after, doesn’t have any way to specify the name of who is generating the response, even on Chat Completions.

You can make sender prefixes to the messages, such as ":vadimigna: " or "[Joe]" in a way that the AI understands. However, the AI won’t completely trust that it was not simply typed by the user.

Responses is not feature-complete to Chat Completions. The only justification for it is to transmit out-of-band messages such as reasoning summaries, and to keep mid-reasoning hosted tool calls from developer inspection.