Memory

Agents remember. Correct something once — "always deploy via the script", "call me Wayne" — and it's saved as durable memory that survives every session. You never have to re-teach the same thing next week, or in next month's conversation.

Memory isn't a black box. It's a plain, browsable list in your workspace: read every entry, edit the ones that drifted, delete what's wrong. Trust comes from being able to see exactly what the crew knows about you.

#What agents remember

Memory is for the facts that outlast a single chat — the things you'd otherwise repeat. Entries fall into a few kinds:

• Facts — how things work here: which server is production, where the build script lives.
• Feedback — your preferences and corrections: "deploy via the script, never by hand."
• Project — durable context about ongoing work.
• Reference — pointers an agent should keep on hand.

Agents save these proactively. When you make a decision, state a preference, or an agent discovers an environment quirk the hard way, it writes that down so nobody has to rediscover it. You can also just tell an agent to remember something.

#Transparent by design

Every memory entry lives in one list you control. Nothing is hidden, and nothing is permanent against your wishes:

• Read it. Browse the full set of what the crew knows.
• Edit it. Fix an entry that's out of date or was captured wrong.
• Delete it. Remove anything inaccurate — it's gone.

#History on every entry

Each entry carries its own history: who changed what, and when. If an agent updated a fact you thought you'd set, you can see the change and roll your understanding back. When two entries seem to disagree, the timestamps tell you which one is fresher.

#Shared across the crew

Memory is shared across the workspace's agents, so the whole team learns together. Teach one agent that releases go through the script, and every agent that touches a release knows it — the correction doesn't have to be repeated agent by agent. Some entries can be scoped to a single agent when a fact only makes sense for one of them, but the default is a shared understanding the crew builds up over time.

Paired with skills — reusable procedures the crew teaches itself — memory is why next month's agents are measurably more useful than today's. The know-how compounds instead of resetting every session.

#Where to go next

• Skills — turn a repeatable procedure into something every agent reuses.
• Agents & your team — who's in the crew that shares this memory.
• Scheduling & automation — put that accumulated know-how to work on its own.