Raven remembers: account-level memory you can open, edit, or delete

Ask an assistant the same question twice in one week. It introduces itself again the second time. Every chat window is a first date. We got tired of that. Raven has memory now, and it behaves like a notebook held by a good assistant.

What makes the cut

Mention once that you fly small planes on weekends, and next month's question about weather minimums comes back answered for a pilot, not for a tourist. That is what Raven keeps: the durable facts. Which project you are on, the stack it runs, the names on your team. The transcript and the small talk stay out of it; what goes in are the notes an assistant would jot down if they planned to see you again.

Internally, the test for what qualifies is literally "would an assistant write this in a notebook". Most of what people say in a chat fails it, which is why the memory list stays short enough to actually read.

You hold the keys

Settings holds the full list of what Raven remembers, every entry visible with what it recorded and when. Deleting is real: a removed memory is tombstoned, so it will not reappear the next time the subject comes up. And say "actually, we moved off Postgres" in conversation; the memory updates on the spot, your version winning over the model's. A memory system that outranks its owner is a design flaw, and nobody here wanted to build that.

It follows you around

Memory rides on the account rather than the device. What you told Raven on the laptop is remembered on your phone and in the web app. It crosses conversations too. The fact you mentioned in April's thread is available in a brand-new chat in November, which means you stop re-uploading context every morning like a stranger reintroducing yourself to the same colleague.

Shared work gets calmer with it as well. Think of the client whose name you always misspell, or the deployment you have been fighting all week. Raven hangs onto the stable parts of that stuff, so every new thread starts warm instead of from zero.

How to keep it honest

When Raven saves something you did not expect, go look at the list. Ten seconds keeps the model's picture of you honest. And correct it in conversation when it is wrong. Say "actually, we moved off Postgres" and the memory updates on the spot. Arguing with your own memory tool until it agrees with you is, we would argue, most of what trust means here.

Open the Raven app, talk to it like someone you will see again tomorrow, and check the memory list whenever you are curious. It is yours, down to the delete button.