[slake]

Sometimes I think we need to expose what 'memories' or their semantic representation has been stored in a 'memory store' so that humans can review and verify it over time. This will help the LLM 'forget' things that the humans using it don't really think is that relevant.