--- title: "How Perplexity's memory works" description: "Perplexity memory explained: what Memory and Brain store, when answers draw on them, and how to view, edit, or delete memories in Settings." canonical: https://past.dev/blog/perplexity-memory date: 2026-09-03 category: Research authors: The past.dev team --- # How Perplexity's memory works Perplexity runs two memory systems. Memory, available broadly, "allows Perplexity to remember details between your conversations" and draws on both saved memories and your search history when it answers ([Memory help article](https://www.perplexity.ai/help-center/en/articles/10968016-memory), updated July 31, 2026). Brain, in Research Preview for Max and Enterprise Max subscribers, builds a structured model of your projects, people, and files inside Perplexity's Computer product ([Brain help article](https://www.perplexity.ai/help-center/en/articles/19700001-what-is-brain)). Perplexity memory is therefore two stores with separate controls, and this page documents both. ## Does Perplexity have memory? Yes, on two layers. The base Memory feature stores "the preferences, interests, and information you share" plus your search history, and consults them "depending on the question and whether using these sources informs a better response" ([Memory](https://www.perplexity.ai/help-center/en/articles/10968016-memory)). When a memory shapes an answer it appears among the answer's cited sources, so you can trace what was used. This is [cross-session memory](/glossary/cross-session-memory) of the same family we documented for [ChatGPT](/blog/chatgpt-memory) and [Gemini](/blog/gemini-memory). Memories accumulate automatically as you ask questions and share preferences. The article's example: ask often about Italian, Chinese, or Mediterranean recipes, and Perplexity applies that preference to later restaurant and recipe requests. Capture favors repetition; Perplexity states the model weighs "the frequency of questions asked," and that it proactively reduces retention of sensitive material such as health and financial details ([Memory](https://www.perplexity.ai/help-center/en/articles/10968016-memory)). Incognito mode is the documented exception: memory and search history are always off there, and incognito questions are never retained or used in memory ([Memory](https://www.perplexity.ai/help-center/en/articles/10968016-memory)). ## The controls in Settings, under Memory The settings page exposes four documented controls ([Memory](https://www.perplexity.ai/help-center/en/articles/10968016-memory)): - "Use search history": whether past searches inform new answers. - "Notes": whether simple details from your sessions are saved for future answers. - "Manage memories": the full memory bank, searchable and filterable, with per-entry deletion. - "Brain" (Max subscribers): the structured system covered below. The toggles are independent, so saved memories and search history can be enabled separately. Two further controls live elsewhere. Whether memory content may improve Perplexity's models is governed by "AI Data Retention" under Settings, then Preferences, then Artificial Intelligence ([Memory](https://www.perplexity.ai/help-center/en/articles/10968016-memory)). And inside [Projects](https://www.perplexity.ai/help-center/en/articles/10352961-what-are-spaces), the owner decides through the "Use session creator's memory" toggle whether sessions also draw on each participant's personal memory ([Memory](https://www.perplexity.ai/help-center/en/articles/10968016-memory)). ## Brain, the second system Brain is "a self-improving memory system that builds a working model of your projects, people, and files," rolling out in Research Preview to Max and Enterprise Max subscribers using Computer ([Brain](https://www.perplexity.ai/help-center/en/articles/19700001-what-is-brain)). It learns in background runs from four inputs: your sessions, your connected tools, files and artifacts you create, and your corrections. Each entry links back to its source session or file, and the store is browsable inside Memory as Concepts, Entities, and Workstreams, where any entry can be edited or deleted ([Brain](https://www.perplexity.ai/help-center/en/articles/19700001-what-is-brain)). Successive runs reinforce entries that remain true, update what changed, and mark stale ones, which is a documented mitigation for [memory staleness](/glossary/memory-staleness). Brain has its own switch under Settings, then Memory settings, then Brain, and organization admins can disable it for members; incognito sessions are excluded from Brain reads and writes ([Brain](https://www.perplexity.ai/help-center/en/articles/19700001-what-is-brain)). In Projects, Brain also maintains a shared, wiki-style project memory that members can read and contributors can edit; Search does not access it ([Memory](https://www.perplexity.ai/help-center/en/articles/10968016-memory)). ## How to delete Perplexity memory Deletion is per-entry or wholesale ([Memory](https://www.perplexity.ai/help-center/en/articles/10968016-memory)): - Open Settings, then Memory, then "Manage memories," and use the trash icon on any entry, or the clear all button to empty the bank. - Deleted memories stop informing future answers. Perplexity "may retain a log of cleared memories for up to 30 days" for safety and debugging. - Turning Memory off stops referencing but "does not clear existing memories"; clearing is a separate action. - To remove a past session's influence on future answers, delete it from your search history. Enterprise accounts behave differently ([Memory for Enterprise Organizations](https://www.perplexity.ai/help-center/en/articles/13654357-memory-for-enterprise-organizations), updated July 24, 2026). Memories belong to the organization rather than the individual, admins manage the feature from the Permissions tab, and disabling it at the organization level deletes all previously generated memories; the same deletion applies when an individual member disables Memory on an Enterprise account. Memory data is encrypted, and Enterprise memory content is never used to train models, per the same page. ## For builders: the same properties, as an API Perplexity's docs describe source-linked entries and stale-entry maintenance, and those are the properties you need again the moment you build an assistant of your own rather than use theirs. past.dev is a memory API for that job: POST /api/v1/ingest accepts raw text with its original timestamp, entity resolution links mentions of the same person across sources, and POST /api/v1/recall returns ranked, dated evidence with sources and one of four status values (Supported, Conflicted, NoKnownSupport, UnknownBecauseDegraded). Facts carry event time, validity windows, and supersession links; deployment is managed or self-hosted in your own Postgres. Start with the [Memory API quickstart](/docs/memory-api/quickstart) and the [benchmarks](/benchmarks).