Handling service session memory in Agents SDK

[dzhouEV]

Hi,

What is the best way to implement something like the customer service example in a FastAPI service? Specifically, what is the best way to handle session memory in the Agents SDK?

Retrieving the traces by group_id seems like one way to do it since that gets stored in the platform but I couldn't find a way to get those programmatically. We could always use other OSS to cache the result.to_input_list() output but wondering if there was a built-in solution.

[rm-openai]

Generally recommend storing the messages in a database. pseudocode:

@app.post(/chat)
def send_msg(message: str, uid:str, chat_id: str):
  previous_messages = db.get(uid, chat_id, count=20)
  previous_messages.append({"role": "user", "content": message})

  result = Runner.run(...)
  db.save(uid, chat_id, result.new_items)
  
  return result.new_items

[tzookb]

thanks @rm-openai

I do notice the result from a run.new_items is a list of

RunItem: TypeAlias = Union[
    MessageOutputItem,
    HandoffCallItem,
    HandoffOutputItem,
    ToolCallItem,
    ToolCallOutputItem,
    ReasoningItem,
]

but for calling run, I need to pass a list of:
list[TResponseInputItem]

ResponseInputItemParam: TypeAlias = Union[
    EasyInputMessageParam,
    Message,
    ResponseOutputMessageParam,
    ResponseFileSearchToolCallParam,
    ResponseComputerToolCallParam,
    ComputerCallOutput,
    ResponseFunctionWebSearchParam,
    ResponseFunctionToolCallParam,
    FunctionCallOutput,
    ResponseReasoningItemParam,
    ItemReference,
]

I want to store data in the DB, but I don't want to blindly store the items. I want to have our own json struct so I can convert to the above objects.
But to which are we aiming for?