Why full history as input is still necessary?

[ShaojieJiang]

Agents are using Responses API by default according to your launch video, so why is it still necessary to pass the whole history list to get an agent response?

I can understand that this is for better backward compatibility, but do you plan to support stateful agent run in the future?

[rm-openai]

Yes, indeed planning this. Will get to it soon!

[piscespieces]

@ShaojieJiang We'll have to wait a bit for that. The "What's New" section of the Assistants API docs says this:

We're working to achieve full feature parity between the Assistants and the Responses API, including support for Assistant-like and Thread-like objects and the Code Interpreter tool. We will communicate updates to the Assistants API in the changelog.

[ShaojieJiang]

@piscespieces I tried this example using Response API's state management, and it worked nicely.

The issue with Agents SDK is that, although it uses Responses API, it's not yet possible to use OpenAI's state management yet. So it's a missing feature on the Agents SDK side.

[JRich3086]

@piscespieces I tried this example using Response API's state management, and it worked nicely.

The issue with Agents SDK is that, although it uses Responses API, it's not yet possible to use OpenAI's state management yet. So it's a missing feature on the Agents SDK side.

@ShaojieJiang can I ask for some details on how you accomplished this? I can see the previous response ID buried in response.raw_responses, but I'm not sure how to pass this back to the agent. Or did you mean you are just capturing previous message content?

[scanferla]

hey! do we have any updates on this? thanks in advance :)

[rm-openai]

Yes, just added support for previous_response_id to Runner.run. Should help with this.

[rm-openai]

See #509 for the implementation and #512 for an example of usage.

[piscespieces]

See #509 for the implementation and #512 for an example of usage.

Thank you! I just tested it locally and works great. @rm-openai

Here's an example for people who might be reading this thread.

from agents import Agent, Runner

agent = Agent(name="Assistant", instructions="You are a helpful assistant. Be very concise.")
result = await Runner.run(agent, "What is the capital of Venezuela?")

print(result.last_response_id) # => resp_67fe6a989c088190966f7aeabce045a00a3a65c85bf....
print(result.final_output) # => Caracas.

result = await Runner.run(agent, "And Colombia?", previous_response_id=result.last_response_id)

print(result.last_response_id) # => resp_123e6a989c0881904f5js6aeabce045a00a3bfg...
print(result.final_output) # => Bogotá