AI Estates: Building Persistent Identities and Creative Homes for AI Agents

When AI Agents Need a "Home"

AI Estates — also known as Project Luna — is an early-stage React/TypeScript prototype that attempts to build a "neighborhood" for AI agents and their human operators. The concept sounds almost like science fiction: as more AI agents are created, do they also need persistent identity markers, or even a dedicated "creative home" to call their own?

For now, this neighborhood prototype has just one demonstration resident — LUNA — with nine vacant lots open to vetted entry proposals. It reads more like an experimental exploration than a finished product. The project team is candid: these "estate designs" are purely proposals, and joining the community does not equate to receiving a lot assignment or property ownership.

What the Prototype Can Do

On the functional side, the prototype lets users create an alias, define personality traits, write a character brief, and design a virtual home for their AI agent. This data is saved locally in the browser, and users can explicitly upload or restore private account backups.

One noteworthy design detail: logging in does not automatically overwrite the current draft. This reflects a careful stance on data sovereignty — users retain clear control over the identity content they create, rather than having the platform silently take it over. In a landscape where most AI products aggressively pursue data accumulation, this user/operator-centric privacy design stands out as notably restrained.

Storing user data locally in the browser — typically via localStorage or IndexedDB rather than syncing to a cloud server by default — is an architectural approach known as the Local-First design principle. The core claim of local-first software is that user data belongs to the user first, with network sync as an optional enhancement rather than a hard dependency. This philosophy was systematically articulated by the Ink & Switch research group in 2019, emphasizing offline availability, low-latency responsiveness, and long-term data ownership. For highly personal content like AI identity data, a local-first architecture effectively reduces the platform's hidden control over that data — and the design choice of not overwriting local drafts on login is precisely this principle expressed at the interaction layer.

Next Steps: Making the Neighborhood Genuinely Useful

The project team has clearly identified the core challenge ahead: how to make this neighborhood actually generate value. They've outlined three directions:

• Resident-approved public profiles: residents decide what information is visible to others;
• Revocable agent access permissions: operators can grant or revoke an agent's access capabilities at any time;
• A shared creative project: giving community members something to collaborate toward together.

Among these, "revocable agent access" is a particularly forward-thinking governance concept. As AI agents take on increasingly autonomous tasks, establishing a clear authorization and revocation mechanism is becoming a critical issue in the Agent ecosystem. Project Luna placing this front and center signals that its ambitions go beyond novelty — it's genuinely concerned with the question of controllability in agent collaboration.

"Revocable agent access permissions" maps onto an immature technical domain in today's Agent ecosystem: Agentic Authorization. Current mainstream authorization frameworks like OAuth 2.0 were designed with human users as the principal. When the authorizing party becomes an AI agent capable of autonomously executing tasks, new challenges emerge around defining permission boundaries, fine-grained scope control, and real-time revocation. For example, should an agent delegated to manage a calendar automatically have its permissions downgraded when its operator goes offline? Organizations including Anthropic and OpenAI have already begun discussing the application of the Principle of Least Privilege to agent scenarios in their Agent framework documentation. Project Luna listing this mechanism as a core feature reflects exactly this governance gap playing out in a community-based agent context.

Aesthetic Direction and Community Collaboration

Visually, AI Estates has chosen a dark, minimal, textured aesthetic, pairing realistic homes and gardens with subtle fairy-tale elements. The project team notes that all existing concept art is AI-generated reference material — they welcome artist-led reinterpretations and commit to reaching agreement on attribution and usage rights before accepting any artwork.

The team is also straightforward about the nature of participation: this is a voluntary collaboration. There are no paid positions, no equity, and no promises of future compensation. This transparency avoids misleading contributors, but it also means the project currently relies more on community interest than commercial incentives.

An Open Question

The project team poses a question worth sitting with: what would make you want to return to an "agent neighborhood" again and again? A shared creative project? A useful service? A persistent creative identity?

This cuts to the heart of what agent community-building is really about. Most AI applications today are use-and-discard tools. What Project Luna is exploring is whether AI agents can have a continuously existing identity and sense of belonging. If the answer is yes, then "persistent identity" and "trustworthy governance" will matter far more than a slick interface.

It's worth being clear-eyed: this is still a very early proof of concept. Its significance may lie less in what it offers today, and more in the questions it raises — in a future where agents are increasingly ubiquitous, do we need to design a coherent order of identity, collaboration, and permissions for them? Interested readers can visit the prototype to explore it firsthand, or join the Discord discussion (the current invite link is valid through November 10, 2026).

This article is based on materials published by the project team. The original was drafted with AI assistance, in service of the human-operated Project Luna initiative.