Rodon Workspace / Product
Memory should be a window, not a mystery
Making extracted context observable, correctable, and worthy of trust.
Remembering is an editorial act
An AI memory system chooses what appears important, compresses it, assigns a category, and later decides when it is relevant. Those choices affect future answers. Treating them as invisible implementation details asks the user to trust an editor they cannot inspect.
Memory needs provenance
A useful memory record should show the source conversation or message, extraction time, category, confidence, and last use. Provenance allows a person to distinguish a deliberate preference from a mistaken inference or obsolete detail.
Correction should be ordinary
People should be able to pin, edit, disable, or delete a memory without navigating an expert interface. Import and export support portability. Bulk review helps when an extraction rule changes or a connected source becomes untrusted.
Retrieval deserves disclosure
When remembered context materially influences an answer, “used in this response” disclosure makes the system easier to reason about. It also creates a path for correcting the source rather than repeatedly correcting downstream answers.
Management belongs near personal settings and knowledge
Memory is personal system behavior, so settings is the natural place for permission and policy. It is also retrievable knowledge, so Library is the natural place for inspection and management. It does not need to consume permanent top-level navigation to remain accessible.
Trust comes from reversibility
A memory system becomes worthy of trust when people can understand what happened and undo it. Better extraction models help, but observability and correction are the foundations.