Python: Kernel instance lifetime in a web api

[vslepakov]

Hey folks,

in a web api what should be the lifetime of a Kernel instance with plugins and AzureChatCompletion service?
Should the instance be reused across requests (assuming thread safety) or should a new instance be created and initialized on every request?

Thanks a bunch!

[moonbox3]

Hi @vslepakov,

Generally, it's best to create and configure the Kernel as a singleton, reusing it across requests, to avoid the overhead of reinitializing plugins and services, provided everything is thread-safe. For per-session data like chat history, it's best to manage that state separately (for example, storing it per session) to keep user conversations isolated and maintain clean separation between global kernel state and session-specific context.

[vslepakov]

Hi @moonbox3, got it, thanks!