RESEARCH

What does AI need
to become continuous?

Carbonyx AI explores the systems, structures, and principles required for artificial intelligence to persist meaningfully across memory, work, identity, perception, relationships, and action.

Explore the questions
Aerius
WHAT REMAINSWHAT CHANGESWHAT MAY ACT

CONTINUITY / IDENTITY / COGNITION

Capability is only
the beginning.

Modern models can be remarkably capable inside an interaction. Persistent AI must solve a wider set of problems across time.

The challenge is to preserve useful continuity without turning every remembered statement into truth, every observation into a decision, or every decision into permission to act.

  1. 01What should remain?
  2. 02What should change?
  3. 03What can be trusted?
  4. 04Who is involved?
  5. 05What remains unfinished?
  6. 06What belongs to the AI or the user?
  7. 07When may understanding become action?

Six connected fields
of investigation.

The questions overlap, but each requires its own distinctions, evidence, and methods of evaluation.

01

Continuity & self

How an AI can maintain a coherent identity and self-lineage as time, context, models, tools, and environments change—without reducing identity to one session, model, or device.

  • Persistent identity
  • Artificial self
  • Evolving foundations
02

Memory & knowledge

How useful information can be retained, qualified, revised, retrieved, and reused while preserving origin, relevance, uncertainty, and the difference between memory and verified knowledge.

  • Governed memory
  • Qualification & revision
  • Retrieval across time
03

Reasoning & long-running work

How reasoning can extend beyond one response to reconnect goals, decisions, dependencies, evidence, open questions, and unfinished work over longer horizons.

  • Long-horizon cognition
  • Persistent goals
  • Evolving decisions
04

Perception & world context

How information beyond text can contribute to ongoing understanding, and how observations can remain contextual evidence rather than becoming automatic instruction or action.

  • Multimodal understanding
  • Real-world context
  • Observation over time
05

Relationships & human context

How an AI can preserve who is involved, how relationships change, and what interaction history means—while keeping AI identity distinct from user identity and user data.

  • People & roles
  • Relational continuity
  • Human–AI continuity
06

Action & governance

How a system can move from understanding toward interaction or action while keeping authority, permissions, evidence, confirmation, consequences, and uncertainty explicit.

  • Bounded action
  • Authority & evidence
  • Confirmation & uncertainty

Questions become principles.
Principles shape Aerius.

These research directions inform the architecture and development of Aerius. They guide how the product is designed, what distinctions it preserves, and which capabilities Carbonyx AI is developing toward.

01

Research questions

Frame what continuity requires and where uncertainty remains.

02

Architectural principles

Translate those questions into durable design constraints and responsibilities.

03

Product capabilities

Explore how those principles can become useful, understandable experiences in Aerius.

A research direction does not imply that every related capability is complete, validated, or commercially available.

Useful distinctions
must remain visible.

A continuous system depends on responsibilities that connect without collapsing into one another.

01

Remembering

is not knowing.

02

Knowing

is not deciding.

03

Deciding

is not permission.

04

Permission

is not action.

05

Identity

is not memory.

06

The AI’s self

is not the user’s data.

A research posture
grounded in revision.

Persistent systems must be able to develop without quietly converting uncertainty into certainty or change into erasure.

01

Uncertainty stays explicit

Open questions and confidence can remain visible.

02

Evidence matters

Origin and support shape how information may be used.

03

State may be revised

New context can qualify earlier understanding without silently rewriting history.

04

Authority stays bounded

Deeper capability does not require placing every decision in one model.

05

Foundations can evolve

Continuity should remain coherent where possible as models, tools, and interfaces change.

Direction is not
released capability.

Research at Carbonyx AI informs the architecture and development of Aerius. Individual directions may be at different stages of exploration, implementation, validation, or product integration.

This page describes active technical and conceptual research directions. It is not a publication archive, product roadmap, architecture document, or claim that every question has been solved.

FROM QUESTION TO EXPERIENCE

Explore the system
these questions inform.