Independent AI development
The ambition is an AI system with its own relation structure, temporal awareness, controlled memory, continuous interaction, and responsibility boundaries. A language model may be a useful tool inside that system. It should not have to be the whole system.
The ambition
Most AI agents depend on a large language model for nearly every meaningful action. My research asks a different question. Can memory, relations, timing, evidence, interruption, and decision pressure belong to a local system that may consult a model without being defined by it?
The goal is not another prompt layer with a character name. The goal is a durable local reasoning structure that can distinguish temporary processing from verified memory and remain accountable to the person operating it.
Evidence before claims
Local test branches have demonstrated verified relation context recall, identity and current time routing, controlled memory writes, quality gates, interruption paths, and relation calculations across primary, secondary, tertiary, and time axes.
Those are meaningful research results. They are not proof of a finished intelligence, and they are not yet a marketable product. I would rather describe the boundary honestly than turn promising tests into a claim they have not earned.
What makes the work different
The research is built in Rust and uses Redis as a working nervous system after the relation shape is proven. It is not a voice wrapper placed around an external chatbot.
Information is connected through primary, secondary, and tertiary relations instead of being treated as one flat transcript. The meaning of a memory depends on what it relates to, how strongly it relates, and whether it has been verified.
A time axis helps distinguish what happened, what is current, what may be stale, and how one event relates to the next. The system should not treat old knowledge and current evidence as if they were the same thing.
The intended conversation is not a rigid record, wait, answer cycle. It uses a continuing perception stream, a temporal workspace, salience, interruption, and voice output that can yield when the person begins speaking.
Temporary processing should not automatically become durable memory. Verified writes, versioned state, quality gates, and clear ownership are part of the design because a useful mind must also know what it should not trust.
Where it stands
The architecture has working tests and promising local behavior, but continuous natural conversation, reliable live audio, broader reasoning, long term stability, and a complete product interface require much more development and testing.
I will not call laboratory progress a commercial release. The work needs the time to become dependable before anyone should be asked to rely on it.
What the next stage needs
This research takes sustained concentration. It needs careful Rust development, repeatable tests, local hardware, voice and memory experiments, documentation, security review, and enough uninterrupted time to follow the architecture where the evidence leads.
Rushing it into another ordinary agent would defeat the reason for building it.
Sponsorship and collaboration
This is an invitation to discuss support for research that is local, experimental, technically serious, and not limited to placing another interface around an existing model.
Support sustained engineering, testing, documentation, and the long periods of attention required to turn a difficult architecture into something reliable.
Contribute toward compute, storage, audio equipment, test hardware, and the practical systems needed to keep the work private and local.
Bring relevant engineering, mathematics, cognitive systems, safety, voice, or product experience to a clear conversation about what can be tested together.
I am not offering a finished product or a guaranteed timetable.
If you understand why local memory, time, relations, interruption, and verified boundaries matter, I would like to hear from you.
Discuss AI development