When do you clean up Serena memories?

[tony-zhelonkin]

How many Serena memories is too many, when you should start thinking about either deleting them and re-initiating Serena on the project as if from ground zero, on an already eveolved project? What is your experience?

Is there reason, instead of wiping them out, compacting them into a single unified memory, e.g. telling Claude to compact them into one? And would there even be benefit from?

I`d like to discuss what are user experiences in how they deal with Serena memories accumulating as the project evolves.

P.S. I know similar questions have already been raised, but couldn`t find how people manage this from their experience

[ketema]

I came up with a trick that is handy for claude code: instruct your agent to activate your serena project in a subagent. This ONLY works if you use a single http server that is persistent. The sub agent context is throw away, but the activated project persists in the running serena http server. Does not work if you are using stdio. Will break if serena http ever supports multiple active projects. However if you have a ton of memories that you don't want to clean and use a single serena http instance this will save you many tokens