Managing multi-turn interactions with ASI:One is much simpler when you do not have to resend the entire message history with every request. The /v1/responses endpoint handles conversation tracking on your behalf, reducing redundant context and simplifying your application workflows.

Here is a straightforward guide to utilizing this feature:

Initial Request Generation
Begin by submitting your prompt using the model identifier and your text input.

{
"model": "asi1",
"input": "Summarize the causes of the 1970s oil shocks."
}

Preserving the Identifier
Make note of the response identifier returned by the system. This key enables you to check the status, extend the dialogue, cancel the task, or remove the entry entirely.

Parsing the Results
The generated output is structured as an array containing distinctly categorized elements, allowing assistant messages, internal reasoning, and tool calls to be processed individually.

Extending the Conversation
When you are ready for the next interaction, you only need to provide the fresh prompt along with the previous identifier.

{
"model": "asi1",
"input": "Now compare that to the 2008 spike.",
"previous_response_id": "resp_..."
}

ASI:One automatically rebuilds the dialogue context behind the scenes, freeing you from passing past turns back and forth.

Choosing an Execution Mode
You can adjust how responses are delivered to fit your architectural needs:
Synchronous processing serves as the default behavior.
Setting "stream": true streams the output in real time.
Setting "background": true begins execution immediately for subsequent retrieval.

Take advantage of stateful messaging to build cleaner, more efficient conversational flows.

Detailed implementation guidance can be found at:
https://docs.asi1.ai/documentation/build-with-asi-one/responses