An architectural layer for AI that holds your full context, mediates disclosure on your terms, and stays portable across providers. Three surfaces, one underlying protocol.
What this is
Every major AI now ships memory. Claude remembers what you told Claude. ChatGPT remembers ChatGPT conversations. Each one knows its own silo. Mandaire is the layer that operates across all of them: twenty years of email, every message thread, calendar history, photos, and AI conversation history from every provider you use. Not what you told an AI. The original sources.
Mandaire exposes that corpus to whichever AI you talk to, under disclosure rules you set, with every claim it makes about you traceable back to the source document it came from. Portable across providers. Inspectable. Correctable. The protocol is documented at mandaire.org.
Three surfaces
The AI Chief of Staff. Holds your full life context across email, calendar, conversations, photos, and the AIs you have already been talking to. Briefs you on what matters at 6:30, drafts in your voice, surfaces the relationship you are about to drop, and refuses the action it shouldn't take.
The AI CTO that refuses to build the wrong thing. Turns your intent into software, records the decisions behind the work, learns your judgment from every correction, refuses the wrong build before it ships. For people with a clear vision and no engineering team.
The canonical specification. The architecture, the bets, the trust commitments, the verification matrix. What Mandaire is, what AIs should do, what we cannot yet prove, and what we have committed to in writing. Open and auditable.
Status
Mandaire is in private beta. The .app and .dev surfaces are live for invited users. The .org protocol page is the canonical specification, kept current as Mandaire develops.