Does Perplexity Comet Remember Between Sessions? What Persists and Where It Stops

Does Perplexity Comet Remember Between Sessions? What Persists and Where It Stops

Short answer: yes, Perplexity added cross-conversation memory, and it reaches into Comet. But it is personalization memory, not automation memory. It remembers who you are and what you prefer. It does not remember what your scheduled agents did, decided, or found last run. For operators running recurring AI workflows, that distinction is the whole story.

What Comet actually remembers

Perplexity's official help center describes Memory as a feature that lets Perplexity remember details between conversations. In practice it covers three things:

Preferences and interests you share. Tell it you prefer concise answers, that you follow the semiconductor market, or that you always want sources cited, and it stores those as memories it can recall in later conversations.

Your search history. Past questions and answers become context the assistant can draw on when a new question would benefit from it, with citations when a memory determines an answer.

Your working context, on the top tier. Max subscribers using Comet's Computer mode also get Brain, which Perplexity describes as a self-improving memory system that turns your sessions into a working model of your world.

All of it is user-controlled: memories are managed in Settings, memory and search history can be toggled independently, and incognito mode disables both. Perplexity calls the cross-model behavior "context portability", switch models and the personal context comes with you, because the memory lives in a store around the model rather than inside it.

That is a real memory system. It is also a memory system built for one job: making the assistant feel like it knows you.

What it does not remember

Now look at it from the operator's chair, where the question is never "does it know my preferences" but "does it know what happened in the last run":

Nothing about scheduled automation state. Comet's guides describe scheduled tasks, daily briefings, recurring checks, timed workflows, as a coming feature, not a shipped one. There is no run today whose state needs to persist into a run tomorrow.

Nothing outside Perplexity's walls. The memory store belongs to your Perplexity account. Your n8n workflow cannot read it, your Make scenario cannot write to it, and there is no documented path for an external agent to reach it over an API or MCP.

Nothing shared. Memory is per-user personalization. Two people on a team running the same research workflow each get their own memories, with no shared operational record of what the workflow itself has learned.

Nothing operational. It remembers that you like concise answers. It does not remember that the 3 AM data-pull run hit a changed API response format and worked around it, or that the Tuesday competitor scan ruled out three vendors. The memory bank holds facts about you, not facts about the work.

This is the asymmetry worth naming: Comet's memory answers "who is this user," while a scheduled agent waking up needs the answer to "what did I learn." Those are different databases, and no personalization feature fills the second one.

The privacy footnote operators should price in

One more thing from the record: Perplexity's CEO has publicly tied Comet's browsing data collection to personalized advertising. The memory feature and the ad model draw from the same behavioral well. Whether or not that bothers you personally, it should shape how you use Comet's memory: it is a consumer personalization store, not infrastructure you would build company automation on. You cannot audit it like a database, version it like state, or export it into your own pipeline.

What your automation actually needs

A scheduled agent does not need to remember your favorite brands. It needs to remember that last run's output, the decisions it made mid-run, the exceptions it handled, and the context the next run should start from. That memory has to live somewhere your whole stack can reach, the browser agent, the workflow runner, the coding agent, all reading and writing one shared store.

That is what Vilix AI provides. It is cloud-hosted with zero infrastructure to manage, so there is nothing to install per machine or per tool. The same memory follows you everywhere over MCP: every agent reads and writes one shared store instead of each tool keeping its own locked memory. It keeps full conversation history, not just distilled facts, so what actually happened in a run is there without anyone writing it down. It is free forever, with a 7-day Pro trial that needs no credit card, and you can export or delete everything anytime.

Comet remembering your preferences is genuinely useful for research. Just do not mistake it for a memory layer. When the scheduled tasks arrive, they will still wake up knowing who you are and nothing about the work, unless you give them somewhere else to remember.