A source on every memory
Each memory keeps a link to the session it came from, so an answer can show its basis.
MokenLogic Memory / Model
In the Mac app · API plannedMokenLogic Memory learns from each person’s finished sessions and keeps what should last: people, projects, preferences, decisions, and how they change. When an AI needs context, it returns a short list of relevant memories, each with its source.
What it keeps / 01
Each memory keeps a link to the session it came from, so an answer can show its basis.
When something changes, the newer fact becomes current. The earlier one stays in the history instead of being overwritten.
Memories that were worked out, rather than recorded directly, are labeled so they can be weighed accordingly.
How it answers / 02
A request receives the few memories that bear on it, not the whole history.
Questions about how many or how much are answered from recorded facts, not estimated.
If memory holds nothing, or only part of an answer, it says so instead of guessing.
Recall is fast and works the same whichever model is in use.
Why a separate model / 03
Personal context should not be locked inside one assistant or one model provider. MokenLogic Memory sits beside the model, so whichever model handles a task receives the same sourced context.
A conversation or task finishes.
Memory updates regularly in the background from finished sessions.
A later request retrieves relevant, sourced memories.
The model handling the task uses them to respond.
Availability / 04
Available in the macOS preview as Episodic Memory. Optional and off by default.
Planned. No public availability date.
In preparation. Results will be published when others can reproduce them.
MokenLogic Memory
Try memory in the Mac preview, or register interest in API access.