Perplexity memory
Perplexity remembers your preferences and past searches, and on Max its Brain builds a working model of your projects for Computer.
by Perplexity · Built this? Edit this page on GitHub
In short
As you search and chat, Perplexity saves short notes about you, like the kinds of recipes you ask about, and can also look back at your past questions and answers. It uses them only when they'd help, and shows any memory it used among the answer's sources. You can switch notes and search history off separately, and search, filter or delete any memory. On Max, a second system called Brain works in the background for Perplexity Computer, the agent that runs tasks. It reads your sessions, connected apps and files, and keeps a linked wiki of your projects, people and open work, updating what's changed and marking what's stale.
Who it’s for
Anyone using Perplexity; no plan limit is stated for Memory. Brain is only for Max and Enterprise Max subscribers using Computer, in Research Preview and still rolling out. On Enterprise, admins control memory for the organization.
What it remembers
Your preferences, interests and details you share, and your past searches. With Brain: projects, people, documents, decisions and open loops, drawn from Computer sessions, connected tools, files and your corrections.
- Memory lives
- Perplexity's cloud
- API keys
- None: sign in to Perplexity
- Works with
- Perplexity, Perplexity Computer (Brain)
- License
- Closed sourceProprietary
- Price
- Plans for search memory aren't stated; Brain needs Max ($200/mo) and is free during its preview
Under the hood
- Who writes memories
- Perplexity, as you ask questions or share preferences. On Max, Brain also writes in a background pass over your Computer sessions, connected tools, files and corrections.
- Where they're kept
- On Perplexity's servers: a bank of memories with a date, category and summary, your search history, and for Brain a linked wiki of concepts, entities and workstreams.
- How they're found
- Memory and past searches are pulled in when the question would benefit, and cited as sources. Computer reuses what Brain knows instead of rebuilding context each time.
- Extra AI calls
- Not published for search memory. Brain is free during its preview, but in a Project, automatic Brain updates use the owner's Computer credits.
How it works
- 1You searchPerplexity notes your preferences and keeps your search history.
- 2Next answerWhen it helps, it draws on memories and past searches and cites them.
- 3Brain (Max)In the background, it turns Computer work into a linked wiki of your projects.
- 4You reviewSearch, edit or delete memories in Settings > Memory.
What stands out
- Memory used in an answer shows up in its sources, so you can see what shaped it and delete that memory.
- Brain's wiki of concepts, people and workstreams, where every entry links back to the session, file or app it came from, and your edits feed its next run.
- Separate switches for notes, search history and Brain, and incognito sessions are never read or remembered by either memory.
What it doesn’t do
- Most of the depth is Brain, and Brain is only for Max ($200/mo) and Enterprise Max, only in Computer, and still a preview. Ordinary search never uses Brain.
- For search memory, how memories are chosen, linked or kept current isn't documented beyond favouring what you ask about often.
- Nothing documented fades with disuse or gets stronger when it's recalled.
- Brain doesn't message you or bring things up on its own. Reminders are scheduled tasks in Computer that you set up.
- Turning Memory off doesn't clear what's saved, and deleted memories may sit in a log for up to 30 days.
- It lives inside Perplexity, on its servers. There's no documented way to run it yourself.
Measured against human memory
1 built, 7 partly, 3 not known of the Field Guide’s 11 mechanisms
Encoding
Salience
Decides what's worth keeping, and how much, as a memory is written.
Perplexity stores key memories as you go, favouring what you ask about often, and tries to keep out sensitive details like health or financial information. There's no documented importance score. source
Emotion
Gives memories written with feeling more weight.
Perplexity doesn't say whether feeling in a conversation changes what's kept.
Storage
Forgetting
Lets unused memories fade over time on its own.
Fading with disuse isn't documented. Brain marks stale entries, but that's about facts going out of date, not time passing.
Interference
When a new memory closely matches an old one, notices and lets the newer one win.
A background pass updates what's changed and marks what's stale, and your corrections update what it knows, but how conflicts between memories are found, or which one wins, isn't documented. source
Retrieval
Retrieval
Remembering something makes that memory stronger.
Whether recalling a memory makes it stronger isn't documented. Brain “reinforces what's still true” in its background pass, not when a memory is used.
Association
Links memories, so recalling one brings its neighbours.
Brain keeps projects, people, documents and decisions as a connected graph you can browse, but how recall follows those links is only shown by example. source
Prospective
Brings a memory back at the right time without being asked.
Computer can run scheduled tasks and set one-time reminders, but memories don't come back on their own, and Brain doesn't message you. source
Transformation
Consolidation
Reworks memories between sessions: strengthening, merging, reorganising.
Brain runs in the background between tasks, reviewing recent activity and reinforcing, updating or marking stale what it knows, each run building on the last. source
Reconsolidation
Corrects a memory when it's contradicted, and keeps what it used to say.
You can edit or delete any Brain entry and your corrections feed its next run, but keeping what an entry used to say isn't documented. source
Gist
Turns many specific moments into general knowledge.
Brain files what it learns from sessions, files and app updates as concepts, entities and workstreams, but drawing general lessons from many moments isn't described. source
Schemas
Forms beliefs about people and projects, and weighs new facts against them.
Brain turns your sessions into a working model of your world, and each Project has its own shared Brain, but how new facts are weighed against it isn't documented. source
Not every system is trying to be a human memory, so a hollow dot isn’t a flaw. It just says what the system doesn’t do. What the mechanisms are.
Perplexity memory’s makers haven’t shared what they’re building next, so only today shows here. Makers can add their roadmap the same way they correct an entry.
Compare Perplexity memory with the others, side by sideQuestions
- Who is Perplexity memory for?
Anyone using Perplexity; no plan limit is stated for Memory. Brain is only for Max and Enterprise Max subscribers using Computer, in Research Preview and still rolling out. On Enterprise, admins control memory for the organization.
- Where does Perplexity memory keep its memories?
Perplexity's cloud. On Perplexity's servers: a bank of memories with a date, category and summary, your search history, and for Brain a linked wiki of concepts, entities and workstreams.
- Does Perplexity memory forget?
Forgetting: not known. Fading with disuse isn't documented. Brain marks stale entries, but that's about facts going out of date, not time passing.
- Does Perplexity memory correct a memory when it turns out to be wrong?
Reconsolidation: a little (Brain (Max, preview)). You can edit or delete any Brain entry and your corrections feed its next run, but keeping what an entry used to say isn't documented. source
- Is Perplexity memory open source?
No, it's closed source. License: Proprietary.
- Does Perplexity memory need an API key?
API keys: None: sign in to Perplexity.
- How much does Perplexity memory cost?
Price: Plans for search memory aren't stated; Brain needs Max ($200/mo) and is free during its preview.