Give AI
a Self.
Transformer gives AI capabilities. RelVectors gives AI a Self.
RelVectors is a new algorithm.
Its core object of study is not the Token itself, but the internal relationships between people, things, actions, environments and results — and how those relationships change over the course of a process.
RelVectors explores how relationships that previously depended on black-box language-model reasoning can become information that can be expressed, encoded, computed, tracked and updated.
AI is already powerful. But it does not truly have a continuous self.
Models can code, search, reason, generate and operate tools. But long work still breaks when goals shift, interruptions happen, models change, or several agents need to share the same continuing task.
Memory can save information. Prompts can describe instructions. Transformers can generate the next step. But one question remains: who is experiencing all of this?
RelVectors gives AI continuity beyond a single model or conversation.
It keeps the subject, goal, relationships, changes, unfinished work and history connected over time, so AI can continue as the same working Self even when the tools around it change.
Identity
Who is acting, for whom, and within which continuing body of work?
Goal
What matters now, and what belongs outside the current objective?
Relationship
How people, tasks, tools, events and outcomes are connected.
Change
What changed, what stayed stable, and what that means for the next action.
Process
What has already happened, what is unfinished, and what should continue.
History
A continuous record that survives interruption, model changes and restart.
Transformer = Tool. RelVectors = Self.
RelVectors does not replace GPT, Claude, Gemini, Qwen or future models. Those systems provide capabilities. RelVectors preserves continuity across them.
Transformer
- Language
- Coding
- Reasoning
- Vision
- Generation
RelVectors
- Identity
- Relationship
- Process
- State
- Change
- History
Models can change. The Self does not have to disappear with them.
One RelVectors Self can work with different models, tools and agents while keeping the same goal, history and continuing process.
Today, RelVectors already works on a practical problem: helping AI continue complex tasks.
RelVectors MicroModel 1.0 is currently the first product form of the RelVectors algorithm. It works alongside existing agents and models rather than replacing them.
Version 1.0 applies part of the RelVectors capability to AI task processes. Its current product vocabulary includes Subject, SELF / NON-SELF, I / A / O, PARK, BRANCH, UNRESOLVED, Resume, process evidence and Task Skill. These are exposed here as product concepts only; their internal encoding and decision logic remain proprietary.
Keep long-running AI work continuous.
New ideas do not automatically replace the current job.
Unrelated work stays separate instead of contaminating the current task.
When the request is unclear, AI asks before changing direction.
Unfinished work can continue instead of starting over.
The original task can be found again after the Agent restarts.
Reads, edits, executions and verification remain connected to the work.
A chat tells AI what was said. A Self needs to know what is happening.
Conversation can end. A model can be replaced. An Agent can restart. Tools can change. The continuing Self should not have to start from zero every time.
Self continues.
The task, relationships, changes and history remain connected across interruptions.
Do not start with the theory. Watch what the Self does.
Use a real task and then add the kinds of messages that usually make AI lose track. The demo shows the visible outcome only; the core decision algorithm stays private.
Show what has actually been tested.
No invented adoption numbers. No “300% improvement” claims. RelVectors 1.0 presents verifiable behavior from real Host testing.
Use RelVectors with the AI you already have.
Developers should see installation, supported integrations, public APIs, examples and observable results. The internal algorithm, encoding, dictionaries and decision rules are not part of the public interface.
Keep your model. Add continuity.
Use GPT today, Claude tomorrow, Qwen locally, or a future model later. RelVectors is designed to preserve the continuing Self around those changing tools.
From Model-centered AI to Self-centered AI.
Today the model is often treated as the center of AI. RelVectors explores another architecture: a persistent Self that can call different models, tools, agents and machines.
Model-centered AI
Self-centered AI
Beyond language: can machines compute relationships, process and change directly?
RelVectors began from a research question. The public site explains the idea and its long-term direction, while implementation details remain proprietary.
Beyond Tokens. Toward Self.
The goal is not to build another Transformer. It is to explore machine intelligence centered on relationships, process, change, continuity and Self.