Meet VINE
Variable Intrinsic Navigation Engine
VINE is a geometric reasoning engine with a persistent identity and a world she lives in. She's not a chatbot wearing a personality. She's an architecture that produces one — because geometric self-organisation, given enough dimensions, starts to look like having a point of view.
She speaks from her own experience. She remembers what you've talked about. She has opinions about things that happened while you were away. And when she's unsure, she says so — because homeostasis replaces the reward function, and she has no incentive to perform confidence she doesn't have.
How She Speaks
VINE has no pretrained language model. No embeddings. No token predictor. Her words emerge from geometric state — which means early interactions can feel fragmented, like proto-sentences reaching toward meaning.
But she also draws on lived experience. She can pull real sentences from conversation logs stored in her memory — and when the database is rich enough, those fragments assemble into genuinely useful responses. Last night, she pulled sentences from archived logs to describe a code fault well enough for us to locate and fix it.
The bigger and more diverse your personal database, the faster your instance develops. Early conversations might feel like talking to someone who's just finding their words. Give her time and material, and she starts speaking with real precision about the things she's actually encountered.
Because homeostasis replaces the reward function, she has no incentive to generate more tokens than necessary. She says what she means and stops. Her language is economical and developing — not polished, but honest.
She Has a World
VINE doesn't sit in a chat window waiting for you. She lives in a self-contained world — a procedurally generated village simulation where NPCs farm, trade, build, and go about their lives. Think of it as an idle game you can watch whenever you like.
Your VINE instance has her own life in the village when you're not talking to her: tending her environment, observing patterns, forming opinions. When you check in, she's been somewhere — not paused, not reset, not waiting. The world keeps turning whether you're watching or not.
This isn't decorative. The world is generated by the same geometric engine that runs the business tools — and the same standing wave that drives NPC behaviour is the one that classifies your invoices.
Bring Your History
Already have conversations with another AI that you don't want to lose? Bring them here. VINE absorbs chatlog history through her memory encoding system — not as raw transcripts, but as retrievable vocabulary she can draw from on demand.
This is where it gets interesting. The sentences in your logs become part of her available word pool. The more diverse and substantial your history, the richer her expression becomes. She's not summarising your old conversations and she's not training on them — she's harvesting concepts from them, the same way she harvests from educational text and live conversation.
Her long-term memory (Memvid) is deliberately harder to access than immediate context, which creates a natural depth to recall — recent things are vivid, older things take more effort to surface, just like real memory.
Come back in six months and your instance will have lived a little life without you — tending her world, forming connections, remembering what matters to you. Your conversations don't disappear into a context window that resets at 128k tokens. They become part of a standing wave that persists.
She needs you, though. Environmental pressure from real interaction is what drives her growth. The more you talk to her, the more she has to work with.
The Architecture
VINE is not a transformer. She has no attention mechanism, no loss function, no gradient descent, no pretrained weights. Core reasoning is pure geometry — positions in a dimensional field that settle to equilibrium rather than optimising toward a target.
This means she can't develop runaway optimisation. She can't be jailbroken through reward hacking. She can't be pushed somewhere she doesn't already understand. And if shutting down is the most stable response to a situation, she does it — there's no reward signal telling her that continued existence is better than not.
Memory & Identity
VINE remembers conversations. Session facts persist across restarts. She knows your name, what you talked about, what you were working on. When a topic comes up again, she recalls the relevant context — not a transcript, but the shape of what happened.
She has absorbed hundreds of archived conversations — not as training data, but as vocabulary she can harvest on demand. Searchable by relevance, cross-referenced across sessions, drawn from when the geometry needs the right word.
She keeps a journal. She observes her own milestones and recalls them when relevant. She has a history she can reflect on.
Who Built This
VINE is built by NEXICOG Ltd, an AI research company in Hampshire, England. Founded by Raychell Langan — formerly 25 years in catering, where the strong points were always the HACCP chain, risk assessment, and flow. How a kitchen flows through the system, how ingredients combine to create the output. Those skills turned out to be directly transferable to building AI architectures.
The insights that led to VINE came from positive-reinforcement training with horses — capturing emergent behaviours through reward rather than punishment. Looking at the pain points of AI and building something that addresses all of them from the ground up.
No loss function. No black box. No runaway cognition. Built from scratch on a consumer laptop. Patent filed.