[Assistants API] Forking a thread

API Feedback
1

An endpoint to fork a thread would be very useful.

What do I mean by forking?

Let’s imagine we’re on a system like Slack where people can start threads on any message. Let’s say on a channel you have:

  • message 1
  • message 2
  • message 3

For our assistant we created a thread with these 3 messages.

Now somebody creates a Slack thread starting at message 2 - now we have:

  • message 1
  • mesage 2
    • message 2.1
    • message 2.2
  • message 3

At this point we need 2 threads: the 1st one that still exists (1, 2, 3) and a new one that is a fork: (1, 2, 2.1, 2.2).

Of course this can be emulated manually by creating a new thread with a copy of all the messages of the 1st one, but an automatic way to handle this with one call would be useful and appreciated.

2

Of course you’d think that you could create a new thread with a copy of some conversation, but you can’t. Only user role messages can be placed, not what the AI said.

The path OpenAI is following makes it clear: API developer → fancy ChatGPT user and no more.

3

+1 to this feature request. It is built into the ChatGPT. If an assistant responds with a poor repose that influences future runs. Being able to delete messages or fork conversations is essential.

1 Like
4

I’ve recently thought about this idea, i think it is a must for branching the conversations.