Skip to content
MantleMantle docs

How Mantle remembers

Everything you add to Mantle becomes memory on its own: you add it, and the brain reads it, indexes it and learns from it. There is no index button.

What happens when something arrives

An email lands, you write a note, you drop in a file or the assistant saves something. The brain then:

  1. Summarises it in a line or two.
  2. Makes it searchable by meaning, so “church work” finds the sermon notes even when they never use the word “church”.
  3. Pulls out facts, such as “the car’s licence renews in March”. A newer fact updates or retires an old one.
  4. Links people, places and projects into a knowledge graph, so it can follow “who works with whom”.

Photos and scans are read too (text in an image is recognised), and so are PDFs, Word files and spreadsheets.

The layers it answers from

Layer What it holds
Persona Who the assistant is, and what it has learned about how you like replies
Journal Who you are, in your own words, carried into every conversation
Recent turns The last stretch of the conversation, word for word
Digests Older conversations, compressed into summaries
Facts Short, current statements about you and your world
Content index A summary and a meaning fingerprint for every item you own
Content The items themselves: emails, files, notes, pages

When you ask something, the assistant takes the slice of each layer that fits the question and answers from it.

It shows its sources

Every fact remembers the item it came from. The assistant can point to the email or file behind an answer. Edit or delete the source and the memory follows.

Example

You forward an email: “Your flight to Lisbon on 14 May leaves at 09:40.” Weeks later you ask, “When do I fly to Lisbon?” The assistant finds the fact, answers “14 May at 09:40” and links the email.

Next