Feature request: Ability to merge flows

[xinha-sh]

I am building a chat tool for my app, a discovery platform for shopify stores.

When the user send messages I am considering the flow to be:

1. Classify the intent - product recommendation or order enquiry
2. If product recommendation is the intent
   • I want to first understand if the product needs to be catered towards men or women
   • based on this I want to find if particular category, subcategory is to be fetched -> custom tool that hits my endpoint to get this
   • use above results to finally fetch matching products -> custom tool that hits an endpoint to fetch recommendation
3. If order enquiry is the intent then
   • ....

At present I am stuck at 2. I can definitely think that it will be sequence of agent calls. But how do I merge this into the original oneOf call.

Current implementaion

const classifyRequestAgent = agent({
  model: openai('gpt-4o'),
  system:
    'You are helpful assistant. You will be given user request and you will classify it as product recommendation or order enquiry.',
});

/*
* Ideally I would like something like
*

  const productRecommendationAgent = sequence([{
    agent: 'determineGenderAgent',
    input: 'Classify the product recommendation request is meant for male or female',
  }, {
    agent: 'populateInputAgent',
    input: 'Populate the missing details in the product recommendation request',
  }, {
    agent: 'productRecommendationAgent',
    input: 'Find the top 3 matching products from Peppyhop',
  }])

 const determineGenderAgent = agent({
   model: openai('gpt-4o'),
   system:
    'You are helpful assistant. You will be given user request and you will classify it as product recommendation or order enquiry.',
 });

 const populateInputAgent = agent({
   model: openai('gpt-4o'),
   system:
     'You are helpful assistant. You will be given user request and you will populate the missing details in the product recommendation request.',
   tools: {populateInputTool},
 })
*/

const productRecommendationAgent = agent({
  model: openai('gpt-4o'),
  system:
    'You are helpful product search agent. You will be given search parameter and you will return matching products from the PeppyHop API.',
  tools: {
    productFetchTool,
  },
});

const orderEnquiryAgent = agent({
  model: openai('gpt-4o'),
  system: 'You are helpful assistant. You will be given user request and you will respond to it.',
});

const processUserChatFlow = sequence([
  {
    agent: 'classifyRequestAgent',
    input: 'Classify the request as product recommendation or order enquiry',
  },
  oneOf([
    {
      when: 'You are asked to recommend products',
      input: {
        agent: 'productRecommendationAgent',
        input: 'Find the top 3 matching products from Peppyhop',
      },
    },
    {
      when: 'You are asked to answer order enquiry',
      input: {
        agent: 'orderEnquiryAgent',
        input: 'Answer the order enquiry',
      },
    },
  ]),
]);