Skip to main content

Memory

Memory settings define how an agent turns raw conversation history into something durable. They come in two kinds, and each kind has its own page in the left sidebar.

PageWhat it configures
SummariesSummary generation: how an ongoing conversation is condensed into a rolling summary, with a prompt and a token limit.
FactsFact extraction: the rules and categories used to pull out and store important facts about your users from their conversations.

A summary keeps an agent coherent over a long exchange by compressing earlier turns into a compact summary that preserves what is needed to continue. Fact extraction is more selective: it identifies durable pieces of information (a preference, a profile detail, an event) and stores them so the agent can recall them later, even in a new conversation.

How the two pages relate

The Summaries and Facts pages are independent settings libraries. On each page you create and manage definitions, and those definitions become available for agents to use through the agent's Memory configuration. An agent can draw on a summary definition, on one or more fact extraction definitions, or on both.

info

These pages let you author and manage the settings. Choosing which summary and fact definitions an agent actually uses happens in the agent's own Memory configuration, not here.

Default settings

Protean AI ships a set of platform-provided default settings so memory works out of the box. Defaults are view-only: they expose a View action (the eye icon) so you can inspect the prompt and configuration, but they cannot be edited or deleted. To use different behaviour, create your own setting alongside the defaults.

The bundled defaults are:

  • Summary generation: protean-generic-summary-generation
  • Fact extraction: protean-behavior-extraction, protean-event-extraction, protean-profile-extraction, protean-skill-extraction

An administrator selects which of these settings act as the platform defaults under Administration, Default Memory Settings.