Docs

Agent memory

plori agents keep notes that survive between conversations. How memory works, private vs account scope, and how to manage it.

A plori agent remembers things between conversations. Memory is a set of short notes the agent reads at the start of every turn, separate from its disk: the disk holds its work, memory holds what it has learned about you and your projects.

How it works

Each note is a short piece of text ("deploys go through the staging cluster first", "the user prefers TypeScript"). Before every turn, the agent's visible notes are handed to it alongside your message, so it acts on them without being asked. Notes are deliberately small; memory is for durable facts and preferences, not documents. Put documents on the disk.

The agent writes its own notes as it works, and you can write notes for it yourself.

Private and account scope

Every note has one of two scopes:

  • Private: visible only to the one agent that owns it. Notes the agent writes for itself are always private.
  • Account: visible to all agents on your account. Use this for facts every agent should know, like house style or standing constraints.

An agent can never share a note on its own. Promoting a note to account scope is a human action in the dashboard, with a confirmation that shows the exact text you are about to share. That keeps one agent's context from quietly leaking into another's.

Managing memory

Each agent's dashboard page has a Memory panel where you can:

  • Create and edit notes by hand.
  • Promote a private note to account scope, or demote it back.
  • Duplicate a shared note into a private copy for one agent to evolve.
  • Quarantine a note to stop the agent from seeing it without deleting it, and restore it later. Useful when you suspect a note is steering the agent wrong.
  • Delete notes permanently.

There is also an account-wide view of all shared notes in the dashboard.

What memory is not

Memory is not a transcript. Conversations are stored with the agent as threads you can reopen, but the agent does not reread old threads on every turn. If something from a past conversation matters durably, it belongs in a note, and a well-behaved agent will put it there itself.