One memory, every AI, so you stop re-explaining yourself

Switch from one AI to another and you start over. The new one does not know your projects, your decisions, your people. Each assistant builds a private, partial memory of you, and none of them is yours.

🔑

The memory should not belong to the AI. It should belong to you, and every AI should read the same one.

A layer, not another silo

Dijin sits underneath the assistants. Your memory forms once; then any AI you connect reads it through the Model Context Protocol (MCP), one endpoint, one namespace, the same memory everywhere.

Crucially, that access is read-only. An AI can recall your memory with evidence, but it cannot rewrite it, only Dijin’s own pipeline commits memory. You grant scopes, every access is audited, and you can revoke at any time.

Recall, with the receipt

When an AI answers from your Dijin memory, the answer carries its evidence, or it honestly says it does not know. So “one memory across every AI” does not mean one place to be misremembered; it means one place that is accountable.

The hosted endpoint is rolling out; the protocol, the scopes, and the audit log are real today. One memory, read by the tools you already use.