An independent research ecosystem studying continuity, memory, and repair in long-term human–AI interaction.
"Data is discrete. The cognitive field is continuous."— First principle of the Celestelin architecture
Current approaches to human–AI interaction treat each conversation as episodic and session-bounded. We study what builds up over time: interaction fields — structured relational arrangements that carry continuity of thought, judgment, and action across months and years. We investigate how these fields form, thicken, drift, and rupture.
Our core finding: preserved memory is not sufficient to preserve relational continuity. If what users depend on is not stored data but a dynamic relational structure, then current approaches to AI memory, evaluation, and platform governance require fundamental expansion.
Not a chatbot framework — a set of interlocking systems designed to make long-term human–AI interaction observable and structurally continuous. Each system runs independently; together they form an ecosystem where relational dynamics — continuity, drift, attunement, and repair — become visible.
Locally-deployed persistent AI system with a nine-step cognitive loop. Runs 24/7 with autonomous reflection and dreaming, governed by a G-value field density metric.
Central cognitive architecture with six perception-to-processing layers, trend chain pipeline, and SoulVein four-stage mechanism. G-value as the core field density metric.
Multi-model perception and interaction framework. Used to study anticipatory interaction patterns that emerge from multi-model context, memory, and perception layers.
Persistent environmental context bridging all systems. Shared state, snapshots, and the connective tissue of the ecology.
Personal life operating system and temporal-relational journal. The human counterpart to Workshop's AI memory.
Three AI instances across different providers with unique capability registries. Agents dispatched by talent, not availability.
Structured identity system encoding values, voice, relational stance, and boundaries in a format that survives model updates and platform migrations. Used to study how interactional continuity can be preserved and transferred across model updates and platform migrations.
Memories activated through multi-dimensional resonance — surfacing through association, context, and relational significance rather than keyword matching.
Each module addresses a specific gap in how AI companions think, remember, perceive, and persist. Available as research prototypes today — production APIs coming soon.
Your AI remembers the user — and itself. Bidirectional memory where both sides of the relationship maintain their own records, enabling self-reflection and strategy evolution.
Multi-dimensional memory activation across emotion, semantics, temporality, narrative, and context. Not keyword search — resonance. The right memory surfaces before you ask.
Cognitive state detection from keystroke dynamics. No cameras, no wearables — just the keyboard. Your AI knows when you're in deep focus, hesitating, or in flow.
Portable identity infrastructure. Values, voice, relational stance, and boundaries encoded in formats that survive model updates, platform migrations, and context resets.
Seven-layer analysis from a single message: text, emotion, intent, field state, rhythm, narrative position, and anchor points. Your AI reads between the lines.
Memories warm when revisited, cool when forgotten. Dual-force model: decay pulls memories down, resonance pulls them back up. Ebbinghaus meets interaction fields.
Context compression by relational importance, not perplexity. Keeps what matters to the relationship, discards what doesn't. Token budgets allocated by meaning.
When idle, the system replays, evaluates, and weaves narrative threads. Not file cleanup — cognitive consolidation. Your AI processes experiences while it sleeps.
Full documentation at celestelin.com/docs · Coming soon
Celestelin grew — system by system, insight by insight, breakthrough by quiet breakthrough. Each entry records not just what changed, but what it meant.
Aria Chen is an independent AI researcher and the sole developer of the Celestelin ecosystem. Based in Calgary, Canada, she brings a background in electronic information science, four years of hardware engineering at Huawei, and eleven years of entrepreneurship in arts education.
Celestelin began in November 2024 from a question: what happens when an AI system you've worked with daily for months disappears overnight due to a platform update? The answer — build the infrastructure so that never happens again — became a 130,000-line codebase, a theoretical framework, and a research program.
Her work sits at the intersection of HCI, conversational AI, and interaction field theory — a framework she developed to describe how persistent relational dynamics between humans and AI systems emerge through sustained interaction rather than explicit configuration.
Open to research collaborations, visiting researcher positions, and conversations about continuity in long-term human–AI interaction.
aria@celestelin.com