Memory

Without memory, every conversation starts cold — you re-explain who you are and what you care about every time. Catalyst’s memory fixes that. It keeps a short profile always in view, and a larger searchable store it can pull from when something becomes relevant.

Open Memory from the sidebar to see and manage all of it.

Profile: always-on context

Your profile is a short description of you — your role, your preferences, how you like answers — that is included in every conversation automatically. It’s the thing the model should never have to be told twice.

Keep it concise. It’s capped at a couple of thousand characters on purpose: it rides along on every message, so it should hold the durable essentials (“I run a long/short equity fund; prefer concise, data-backed answers”), not a diary. Edits apply to your next session.

Entries: a searchable memory

Beyond the profile, Catalyst keeps entries — individual facts and notes — in a store it can search by meaning, not just keywords. When something in a conversation is relevant, the model recalls the entries that relate to it and works them in:

Recall is ranked so the most relevant and most recent entries surface first, and entries you pin are always available. The model writes new entries as it learns things worth keeping, and you can add, edit, pin, archive, or delete them yourself from the Entries tab.

Summaries: what happened in past sessions

Catalyst can also keep short summaries of your past conversations, so the gist of last week’s work is recallable without replaying the whole transcript. New summaries are created automatically for recent sessions; use Run Sweep to catch up recent ones on demand, or Sync Session to summarize an older conversation that’s outside the automatic window.

Tip

Think of the three tiers by how often they apply: profile is true always, entries are pulled in when relevant, and summaries capture what happened. Together they let Catalyst pick up where you left off instead of starting over.

You’re in control

Everything in memory is visible and editable on the Memory screen — search it, edit it, pin what matters, archive what’s gone stale, or Clear All to wipe it and start fresh. Nothing is hidden: what the model can recall is exactly what you can see.