The problem of starting over every morning
You spend a Tuesday afternoon explaining to Ship that your team prefers Monday releases over Friday ones, that the brand voice avoids exclamation points, that your billing system can't take a customer over a thousand dollars per month without a manual review. Wednesday morning you open the chat and Ship has forgotten all of it.
This was the failure mode that made chat-based AI feel like a clever toy. You spent more time re-explaining than you spent moving forward. We built memory into Ship so that the explaining only happens once.
What "memory" actually means here
Memory in Ship is the catalog of durable facts Ship carries from one chat to the next. Not your message history — that's stored too, but it's not what memory means. Memory is the distilled stuff: preferences, decisions, constraints, the things you said once that you'd want a colleague to remember next week.
When you tell Ship "we prefer to ship on Mondays," that becomes a fact in memory. When you say "the team agreed to standardize on dependency injection in the backend," that becomes a fact. When you say "I'm thinking about the auth flow project right now," that's context Ship can use next time you mention auth.
The next morning, those facts are loaded before the conversation starts. Ship doesn't re-ask. It acts on what it already knows.
How facts get captured
You don't have to do anything. As you talk to Ship, the parts of your messages that contain durable preferences or decisions are extracted into memory automatically. Noise — "ok," "thanks," small acknowledgments — doesn't get saved. Neither do questions or in-flight thinking that hasn't settled. When Ship is uncertain, it leans toward saving. Memory that's wrong is annoying; memory that's missing is the failure we're trying to prevent.
Where you can see and edit memory
Memory surfaces inline in the conversation — when prior facts inform a reply, Ship shows how many it drew on ("3 prior memories"). To correct or remove a fact, tell Ship directly in chat or through your agent ("forget that we ship on Mondays" / "update that note — we ship on Wednesdays now"); Ship adds, and you curate by asking. Memory drifts — a decision you made six months ago about a feature you've since rebuilt is now wrong — so the catalog is yours to curate. (A standalone Memory management page is not part of the current console.)
When memory has the wrong thing
If a fact contradicts what you're saying right now, Ship asks rather than guesses. "I have a note that you preferred Monday releases — is that still right, or is the change permanent?" That's a small interruption, and it's the alternative to Ship silently overwriting a decision behind your back. You answer, the memory updates, the conversation continues.
What other people can see
Memory is personal. Other members of your workspace cannot read your memories. The owner cannot read your memories. If you're a member of two workspaces, your facts in each are kept separate. Your half-formed thoughts about strategy belong to you, even though the decisions that come out of them get written to shared tracker and shared knowledge where the team can see them.
When memory pauses
Sometimes you want a conversation that doesn't go to memory — a hypothetical, an exploratory thread, something private even from yourself. Tell Ship to pause memory for it ("don't save anything from this conversation"), and resume when you're done. Nothing in that thread gets extracted while it's paused.
When memory is the wrong answer
Memory is for durable preferences and decisions, not for facts the team should be able to consult. If a piece of information should be reusable by developers, designers, or future hires, it belongs in workspace knowledge, where the whole team can see it. Memory is personal on purpose; knowledge is shared on purpose. Different drawers.
The next page covers the other thing that makes working with Ship feel like a colleague rather than a tool: switching what you're working on in the middle of a conversation.