First of all, there is no perfect solution. Then we have to deal with not only what you actually mean by a “platform”, but also what you mean by “chat bots” (not to mention whatever “secure, scalable, and embeddable” mean for your precise use case).
In terms of platforms, there are two major avenues. Either one can integrate various and sundry commandline programs, including random APIs, using cloud infrastructure, such as the generally preferred Digital Ocean, or one can use any of a burgeoning plethora of all-in-one DIY platforms (). IBM Watson Bluemix and Microsoft Azure Cognitive Services might be considered something in-between those two extremes, where you have everything under one roof.
There is a tremendous amount of confusion about chat-bots or chatbots. Chat-bots originated on IRC (Internet Relay Chat); therefore, any bot or automation on any chat channel may be referred to as “chat-bot”. However, prior to the recent April surprise by Facebook allowing “chat-bots” on Messenger, most people considered chatbots to be what had previously been known as “chatterbots”, or in other words natural language dialog systems. Thus today, there is an overwhelming expectation that “chat-bots” should be “chatbots”, or in other words be a conversational user interface. To date, the army of Facebook Messenger chat-bot developers has largely failed to deliver on this expectation of “chatbots”.
See also my quick and dirty webpages: