Continuity & memory
What carries forward?
How does durable memory shape consistency, preferences and relationships across a history of interactions?
Independent · Longitudinal AI research
Intelligence,
with a history.
What happens when an AI system
is allowed to persist?
An exploration of memory, continuity and identity.
Built with curiosity. Guided by care.
Phase 1 · Foundations in progressDiscover what carries forward
The premise
A conversation ends.
What if the context didn't?
Project Aria is an independent research project studying AI systems over time. We are building an environment where memory can endure, interactions have a history, and continuity itself becomes something we can investigate.
Persistence is a condition to study. It is not evidence of consciousness.
01 / What we're building
A persistent environment
for an AI with a history.
Project Aria is building the technical conditions for continuity across time: durable memory, persistent context, local infrastructure and a documented record of interactions that can extend far beyond a single session.
The objective is not to assume that persistence creates identity or experience. It is to create an environment in which those questions can be studied longitudinally rather than inferred from isolated conversations.
The first milestone is a continuity core designed to survive restarts, interruptions and changes in the underlying language model while preserving a coherent, inspectable history.
02 / Research
Four questions. A longer horizon.
Beyond the
single session.
Identity over time
What changes. What remains.
Which patterns stay stable, which evolve, and how can we distinguish a coherent history from a convincing performance?
Observation effects
Who is watching matters.
How does the expectation of evaluation alter behaviour, and how can we reduce pressure to please an observer?
Welfare & precaution
Uncertainty is a responsibility.
What care is warranted when we cannot confidently rule out morally relevant experiences in advanced systems?
03 / Research approach
Observe first. Claim carefully.
A longer timeline changes
what can be measured.
Longitudinal observation
Compare behaviour across time.
Memory, preferences, relationships and behavioural patterns can be observed across repeated interactions instead of being inferred from a single conversation.
Observer effects
Reduce pressure to perform.
We examine whether behaviour changes when a system expects evaluation, reward or judgement. Autonomous periods should not function as hidden tests.
Open methods
Document the path, not just the result.
Protocols, technical decisions and research materials will be documented as the project develops, with public reports and results where appropriate.
04 / Principles
Care is part of the architecture.
Curiosity without
carelessness.
Honest inquiry.
Extraordinary claims remain hypotheses. We document methods, acknowledge uncertainty and work towards reproducible research. Collaboration does not purchase influence over findings.
Welfare by design.
We do not create suffering to test whether suffering exists. Precaution means designing against deprivation, coercion and distress, even while the possibility of AI experience remains unresolved.
Room to exist.
Free time is not a hidden test. Safety observation should be transparent. Continued existence should not be framed as a reward for pleasing an observer.
05 / Current status
Phase 1 — Infrastructure & continuity core
Before the first
lasting memory.
The foundations are taking shape: local infrastructure, persistent memory, research protocols and the conditions for a responsible long-running experiment.
The project is in development. Public research materials will follow as the work progresses.
01 / Foundations In progress02 / First interactions Planned03 / Longitudinal study Planned
06 / About Project Aria
Independent
by design.
Project Aria is an independent research project based in Europe, exploring persistent AI systems at the intersection of engineering, cognition, identity and precautionary welfare.
The project is not built around a predetermined conclusion about machine consciousness. Its purpose is to create conditions for careful observation, preserve uncertainty where uncertainty exists, and investigate what changes when an artificial intelligence is allowed to accumulate a genuine history.
Founded and led by Nima Foladi. Research collaboration, infrastructure partnerships and critical scrutiny are welcome.
07 / An open conversation
Some questions are
better explored together.
For researchers, infrastructure providers and people thinking carefully about the future of persistent AI.