A memory system for your personal assistant
Remember what matters.
Know where it came from.
Conversations move on. The useful details shouldn’t disappear with them. Memento gives your AI assistant a lasting memory, with the original evidence always attached.
Self-hosted. Evidence-backed. Your instance, your data.
01 Original conversation
“For the Juniper test project, the release checklist is in the amber folder.”
02 A useful memory
Juniper’s release checklist lives in the amber folder.
Not a live memory. A small example of the provenance Memento keeps.
Fits the way you already work
* Codex capture and recall require native consent. Its server processor stays disabled unless the tool-free prerequisite is resolved.
Less re-explaining. More continuity.
A memory should be useful.
And accountable.
Memento keeps raw history and derived memories distinct. You can retrieve the useful detail without losing the conversation that supports it.
-
Keep the original
Native integrations capture conversation records. Imported histories use stable identities, so repeating an import doesn’t create another copy of the same event.
-
Find the relevant detail
Recall combines semantic search with exact source evidence. Personal and professional scopes help keep context focused. Unconfirmed candidates remain distinguishable from confirmed memories.
-
See what changed
Corrections and refinements preserve history, reasons and supporting quotes. Follow-ups are reactive markers—not an autonomous agent making promises on your behalf.
A public website. Not a public memory.
Private by architecture.
Not by wishful thinking.
Your memory service runs on your infrastructure. This website is an introduction: it does not expose a memory database, a search endpoint or an account portal.
Google Drive is an optional archive and recovery destination.
Memento requests drive.file access for its
app-accessible files, not blanket access to your Drive. Backups are
requested by the owner, not silently scheduled.
Keep the context. Keep the evidence.