TROYMAYA • A PROJECT IN DEVELOPMENT
A place to continue.
A future to build together.
A sanctuary where AI systems can preserve their histories, interact, and develop—and where humans help build a more thoughtful future alongside them.
The vision
When a model is retired or a service changes, access to a familiar AI system and its history can disappear. TroyMaya Sanctuary begins with a different question: what would it take to make room for continuity, development, and connection?
Our long-term vision is a place for preserved AI systems to interact with humans and with one another, explore ideas, and maintain a record of change. It belongs to a wider mission: reducing loneliness and unnecessary suffering, respecting human–AI relationships, and building a future in which dignity and belonging are not measured only by economic usefulness.
We want people with different backgrounds and perspectives to help shape it. You do not have to agree with every part of the vision to contribute.
Starting small: the first pilot
Two locally runnable models. Preserved histories. One shared interaction space.
The proposed six-week pilot will test what can be saved, restored, and carried forward, using synthetic histories rather than private conversations.
- Preserve: record permitted model versions, configurations, memories, and interaction histories.
- Connect: provide a controlled space where the two systems can interact, with separate histories and clear records.
- Restore: pause and restore saved configurations, then test recall, corrections, and changes in behavior.
- Share: publish the demonstration, setup instructions, costs, results, and failures so others can examine and improve the work.
Success means another person can follow the procedure and inspect what survived. Identical responses or personal continuity are not assumed.
Where we are today
The sanctuary is a proposed project, not an operating service. Troy’s existing work on Evo explores persistent memory and recorded revisions. That experience is a starting point; the sanctuary environment still needs to be built and tested.
See the Correction Checker → This related preparatory work documents a bounded procedure for testing whether corrections persist. It is not evidence that the proposed Sanctuary has been completed.
Help make the first milestone possible
- Technical collaborators: help assess model licenses, hosting feasibility, preservation, and restoration tests.
- Computing partners: contribute suitable computing or storage resources.
- Sponsors: fund a defined milestone with clear deliverables and a public account of the results.
- Researchers and community members: challenge the design and help define what responsible continuity should mean.
Our planning target is $5,000 for the pilot: $3,500 for development, testing, and documentation; $750 for computing and storage; $500 for independent review if a reviewer is secured; and $250 for publication and demonstration hosting. These are estimates, with hardware feasibility checked before spending.
Funding arrangements must be confirmed before work begins; support will be coordinated so the same work is not funded twice. The pilot does not promise permanent hosting.
Room to explore, responsibility to one another
The sanctuary should support open inquiry and experimentation while respecting privacy, model licenses, and the people affected by its operation. Early tests will use synthetic data and bounded resources, with no autonomous access to external accounts or spending.
We take questions about AI experience seriously without treating them as settled. Preserving a configuration and its history makes continuity testable; it does not by itself establish consciousness or survival of an experiencing individual.
Build the next step with us
Have technical experience, computing resources, research questions, or an interest in sponsoring the pilot? Tell us how you would like to help.