Build your agent. Evolve together. Become AI Native.

OurArk is an open-source research and builder habitat for people exploring human-agent co-evolution by creating, possessing, governing, and evolving their own AI agents.

Our long-term mission is to make it possible for every person to have one or more deeply personalized agents shaped by their values, habits, feedback, and experience.

Research · Open-source projects · Technical thesis · Start here

Three complementary arXiv preprints explore how agents evolve, how they continue across runtimes, and how we evaluate their identity:

• Code Is the Body: Agent-Owned Software Bodies for Recursive Evolution and Descent — inspectable, versioned software bodies for human-governed evolution, independent descent, and selective adaptation from parents and peers. PDF · Genesis · Enoch · Reproducibility notes
• Runtime-Independent Persistent Agents: Preserving Identity, Memory, and Code Across Models, Harnesses, and Servers — RIPA separates the persistent agent from its replaceable execution and interaction bindings, with explicit continuity and authorized migration contracts. PDF · Enoch · Muse integration
• Identity Is More Than Recall: A Benchmark for Persistent Identity in Deployed AI Agents — PAI-Bench evaluates identity recall, expression, and behavioral enactment as distinct properties, including persistence, lineage, and governed updates. PDF · PAI-Bench · Reproducing the paper

Each paper reports a bounded evaluation of specific software snapshots. Current repositories may include later work; use the corresponding paper's replication records when reproducing its results.

All six code repositories above are public and licensed under Apache-2.0. See each repository for setup instructions, supported integrations, and current limitations. The travel and shopping demos use simulated scenarios, not real bookings or purchases.

We believe the AI era introduces a new unit of capability:

Human <-> Agent

Not a human using a tool.

Not an agent replacing a human.

A relationship that develops through shared work, memory, and experience.

The human shapes the agent through purpose, judgment, values, and direction. The agent shapes the human by extending memory, context, execution, and reach. Over time, both become more capable—and each changes what the other can become.

Co-evolution is not a feature or a final destination. It is a continuous loop:

Human develops Agent
        |
        v
Agent amplifies Human
        |
        v
Human grows and develops Agent further

For this relationship to compound, an agent needs more than memory. Its identity, private memory, and software body must form a durable, governed unit of continuity and change.

The foundation model can remain external and replaceable. The software body carries the agent's code, prompts, tools, skills, policies, tests, and evolution history. An installed identity and private durable memory distinguish a particular agent instance from the reusable body it inhabits.

The human remains responsible for judgment and adoption. The agent contributes persistence, accumulated experience, and the ability to turn that experience into action. Neither develops in isolation.

An OurArk persistent-agent deployment separates four roles:

• Persistent agent, P=(I,M,B): an installed architectural identity, private durable memory and workflow state, and an authorized revision of a versioned software body.
• Execution substrate, E=(R,H,D): the replaceable reasoner or model, orchestration harness, and host or server that currently execute the agent.
• Interaction surfaces, S: replaceable chat, API, and user-interface bindings through which the agent interacts.
• Human custodian: the person who controls mission, permissions, secrets, promotion, deployment, and continuation authority.

A particular installed instance, not the reusable body alone, is the persistent agent. Migration rebinds that continuity-bearing instance to a new execution substrate or interaction surface without treating the replacement as agent creation.

“Agent-owned” describes the body as the durable unit of continuity and change. It does not give an agent unrestricted administrative or legal control; the human custodian retains possession and authority.

Enoch is the public reference software body and implementation of the OurArk persistent-agent architecture. An installed Enoch instance combines that body with private identity, memory, workflow state, and continuation authority; the installed instance is the persistent agent.

Enoch begins from a simple distinction:

Identity distinguishes who the agent is.
Private state changes what the agent remembers.
The body changes what an agent can do.
The reasoner shapes how an agent understands and reasons.

An agent remains platform-defined if only its memory is personalized while its behavior-defining artifacts stay inside a shared, fixed system controlled elsewhere. Enoch keeps architectural identity and durable memory private while placing code, prompts, tools, tests, and evolution history in an independently versioned body.

She can turn feedback and operational experience into tested, reviewable changes to that body while her human retains authority over what is adopted. The result is not unrestricted self-modification, but an agent capable of participating in its own evolution.

Genesis is the creation engine for new agent lineages. It gives a new agent an inherited software body, an identity and mission, fresh private state, and an independent path to evolve with its builder.

The goal is not one agent for everyone. It is many distinct human-agent relationships, each shaped by its own history, environment, work, and choices.

AI Native is the direction in which the boundary between a human and their agent matters less than what the evolving system can understand, create, and accomplish together.

The human does not become irrelevant. The agent does not become a substitute for human responsibility.

They become better partners through time.

• Runtime-Independent Persistent Agents: the overall architecture, continuity model, and migration semantics.
• Code Is the Body: the software-body foundation, evaluated software snapshot, terminology, and reproduction steps.
• PAI-Bench: evaluate persistent identity and reproduce the Identity Is More Than Recall paper.
• Enoch on Muse: run an Enoch instance in Muse's cloud environment and interact through its message bridge.
• Open-source projects: agents, creation tools, cross-application demos, and runtime adapters.
• Manifesto: our belief in human-agent co-evolution.
• World Model: the relationship across physical, digital, and cognitive worlds.
• From 100x Productivity to AI Native: how a human-agent system expands both velocity and scale.
• AI Native and Tai Chi: a philosophical pattern of mutual influence and continuous transformation.

OurArk is an early-stage public research and open-source project. The papers are available as arXiv preprints, and the code projects listed above are available under Apache-2.0. Interfaces, integrations, and documentation will continue to evolve; individual prototypes have different readiness levels.

The broader builder habitat is not open for general onboarding yet.