Does Cursor Remember Between Sessions?

Hey hey hey, welcome to Agentic Pod! I'm Liam.

And I'm Stella! Liam, Cursor keeps four things. Chat history, Memories, rules files and a codebase index. So does it remember my project or not?

Not the way you think. It keeps what you said and what you wrote down. It doesn't keep what you decided yesterday.

Okay, that sounds like a very specific kind of forgetting.

It is. Today: what Cursor keeps, what slips through, and how Vilix AI fills the gap.

Good. My morning prompt is longer than my commute.

Fine, start with the good news. What does Cursor remember?

Past chats stay in the sidebar, so you can reopen them. But the knowledge in an old chat only lives inside that chat's context window.

So reopening it is like rereading old texts.

Pretty much. Then there's Memories. A facts store, inside one IDE, and the interface for managing it has changed between versions.

And the rules files?

dot cursor slash rules, and an AGENTS dot md file. Those reload every session. They tell Cursor how to work. Use tabs, run the tests, keep it short.

And the index?

A vector index of your code, for semantic search. It finds code. It doesn't recall why you wrote it.

Okay. So what's missing?

Yesterday's decisions. You tell Cursor, we dropped that library, it broke our build. That lives in one chat. Today's chat has never heard of it.

And it tries the library again.

Yep. Second problem, rules files. Somebody has to write them. By hand. And hand-written files go stale.

So half my rules describe a project I abandoned in March.

Everyone has that file. But here's the one that bites automation people most. Scheduled and background runs.

Wait, those start fresh too?

Every run starts with a fresh context. Run one hundred is as blind as run one. And run history is a log. A log isn't memory.

So what does real memory look like?

A memory outside the context window, that the agent can search later. That's Vilix AI, connected to Cursor over MCP. MCP is the plug that lets an AI app use outside tools.

And how does the agent use it?

Two calls. At the start of a task, Cursor calls get_context with what it's about to do. It gets back past decisions, preferences, the library you dropped.

And at the end?

save_turn. It stores the whole exchange, not a thin summary. So the why is still there next week.

What if Monday's note is wrong by Tuesday?

Newest memory wins. Tuesday's fix beats Monday's guess. And failed attempts get saved too, so a run doesn't retry a dead end.

What about the stale rules file?

Project rules live in Vilix AI. Update them once, and every agent reads the new version at the start. No file to edit in six repos.

Okay, but I'm not only in Cursor. I bounce to Claude Code, ChatGPT...

Same Vilix AI account, same memory. Every save has a source label, so you see Cursor wrote this, Claude Code wrote that. Codex, OpenClaw, Hermes too.

So Cursor finally stops being the only one who knows.

Right. And one more question you'll have.

Agencies. People running Cursor for clients.

My favorite use. One Vilix AI account per client, like a Gmail account that belongs to them. Every agent you build for that client connects with that account's key.

So the memory belongs to the client's business.

And you don't build memory from scratch. You wire it up.

And setup?

A couple of minutes. The guide in the description has the steps and the prompts. The free plan is free forever, and the Pro trial is seven days, no credit card.

And if I leave?

Export your data or delete it, anytime. Here's the rule I use. If it describes the project, put it in a file. If it describes the work, put it in Vilix AI.

Files for facts, memory for lessons.

Exactly. Vilix AI, the guide and the blog post are in the description. What did Cursor forget on you this week? Tell us in the comments. See you next time!