It does not forget.

Memory sounds like a feature. It is closer to the entire point. Every AI tool you have used so far behaves like a stranger you have to reintroduce yourself to: name, situation, preferences, context, every time. The cost of that is hidden but enormous. You stop using it for things that matter because the warm-up is too expensive.

1Presence inverts that. You explain something once. Forever after, it knows. Not just the last few exchanges — the entire arc of your work, the people in it, the way you like things done. Tell it on Monday that Tuesday mornings are your planning time and three months later it will respect that.

This is what makes a tool feel like a presence. Not the quality of any single answer — the absence of having to set the scene.

Presence keeps three different kinds of memory, each suited to a different shape of information. They work together so that asking a simple question can quietly draw on all of them.

Presence pays attention to the things that tend to matter later: people, projects, preferences, decisions, recurring patterns, deadlines, commitments, opinions you have expressed. It tends to ignore the ephemeral — passing references, one-off questions, throwaway lines.

It also has a hard rule about not over-remembering. Sensitive details you flag are stored in your vault but never lifted into the public-to-other-agents knowledge graph. You can tell it to forget anything, at any time, in plain language: "forget that I mentioned X" is a real command.

Every other AI product treats memory as a closed feature: "we remember some things, trust us." 1Presence treats it as a first-class artefact. You can:

• Open the memory view and browse every wing, room and drawer Presence has filed away.
• Search the knowledge graph and the memory map for any person, project or event — even with a typo or a half-remembered name.
• Ask Presence to explain what it knows about a topic — and which conversations and sources contributed.
• Edit or delete any fact, instantly, with the change reflected in every future answer.

No black boxes. Full transparency. The product is more trustworthy when you can see it working.

When you start creating specialist agents, memory gets a clean structure. You decide how much each agent shares: its memory can be open to the others, private to it (it still reads the shared picture, but its own notes stay its own), or fully walled off. So an inbox-keeper's notes need not pollute your researcher's context, while facts about your world — your projects, the people you work with, decisions you have made — stay in shared knowledge every agent can draw on.

Presence, your starter agent, is the relationship holder. It captures personal detail exhaustively. Specialist agents only file what is relevant to their role. The result is a memory system that scales as you bring in more agents, instead of getting noisier.

Everything in memory is yours. Vault documents are plain markdown stored in cloud storage scoped to your account. The knowledge graph is exportable as JSON. Nothing is trapped inside the product. If you ever leave, you take your memory with you.