Feature request for product/service
Chat
Describe the request
There are many instances when dealing with a large codebases that a task/analysis can proceed on subsets of the file tree independently and in parallel. And then the results aggregated as each completes. Even if not in parallel, resetting the context to a fixed point for each batch would solve quite a few issues with context drift (i.e. agent biased to the recently seen and unrelated code)
Is this something that multi-agents supports or something planned?
Does anyone, in the meantime, have any strategies they use, short of manually duplicating the chat at the fork point?