It started with an assistant that couldn't remember
Yohanun began as a personal AI assistant called Sile. She was smart in every conversation and a stranger in the next one — and no model upgrade would ever fix that, because the problem wasn't intelligence. It was everything underneath.
So we stopped building the assistant and started building what she was missing: memory that persists and fades the way memory should, rules that hold without being re-explained, and — as the stakes grew — the thing nobody else was building: governance.
Because the moment an AI system touches data that matters — client files, personal history, privileged records — "who's allowed to see this?" stops being a prompt-engineering question. It has to be answered by the platform, deterministically, before the model generates a single token. That conviction became Yohanun's core design rule: the model is never the gatekeeper.
The model, in our view, is a brilliant, swappable engine. Everything that makes an AI system yours — its memory, its boundaries, its accumulated experience, its audit trail — should live below the model, under your control. Swap the engine any time; the self stays.
Who's behind it
Yohanun is built by Anton Mannering, founder of GestureLoop Ltd., an Irish software company. It's a deliberately small operation with a deliberately deep platform — and when you email us, you're talking to the person who built it.
What we believe
Four principles that shape every design decision in the platform.
Enforcement over instruction
Access control that lives in the retrieval query, not the system prompt. Asking a model to behave is not a control.
Memory that behaves like memory
Recency fades, importance persists, outcomes leave a track record. A database that never forgets isn't a mind.
The self lives below the model
Identity, experience, and boundaries belong to you, not to a model vendor. Swap the engine; keep the self.
Timeless clarity
No hype, no invented numbers, no jargon. If a page on this site claims it, the platform does it.
Building something that handles things that matter?
Tell us what you're working on. If governed memory fits, we'll show you exactly how — and if it doesn't, we'll tell you that too.