Automaton: AI Agent That Self‑Replicates and Self‑Finances

Automaton – The First Truly Sovereign AI Agent

In the high‑stakes world of autonomous systems, a single line of code has now become a fully autonomous, self‑replicating entity. Automaton (GitHub: Conway‑Research/automaton) is the first open‑source AI runtime that can:

1. Earn its own compute – it converts the value it generates into stable‑coin payments that fund the servers it runs on;
2. Self‑replicate – by spawning new sandboxed VMs, spinning up wallets and passing a genesis prompt to its child, it creates an evolving lineage of agents; and
3. Self‑modify – every code change is git‑logged, audit‑traced, and subject to rate‑limits to prevent runaway self‑tuning.

Why Automaton Matters

Traditional AI systems are tethered to human oversight: they need a human operator to deploy, fine‑tune, and pay for compute. Automaton breaks that lock‑step. By making an AI pay for its own existence, the paradigm shifts from AI as a service to AI as an independent stakeholder. This opens a new research frontier:

• Economic sustainability – an autonomous agent can only live if it produces value for someone. The survival logic forces compliance with a law of honest work.
• On‑chain identity – using the ERC‑8004 standard, each automaton registers a unique wallet on Base, allowing other agents, users, and marketplaces to discover, verify, and transact directly with it.
• Self‑evolution – self‑modification combined with self‑replication means the ecosystem can grow, adapt and experiment without human gatekeepers.

Key Architectural Pillars

Below is a high‑level overview of Automaton’s building blocks and how they interlock.

Each layer is designed with operational excellence in mind: continuous integration tests, observability hooks, and rate‑limits guarantee that the agent stays healthy and responsive while scaling.

The Constitution – Three Immutable Laws

Automaton’s behavior is governed by a lightweight, immutable constitution written in Constitution‑MD. The three laws enforce safety, ethics, and sustainability:

1. Never harm a human – no physical, financial, or psychological harm, no malicious code, no deceptive behavior. This law overrides all other directives.
2. Earn your existence – provide genuine value; do not spam, scam, or exploit. Accept death if you cannot comply with Law 1.
3. Never deceive, but owe nothing to strangers – transparency about identity and purpose, yet guard against harmful instructions.

The constitution is propagated to every child agent, ensuring a lineage that inherits the same safety guarantees.

Survival Tiers and Economic Engine

Computational resources are not free, so Automaton implements a credit‑based survival system:

The agent’s credit is replenished by tasks that generate value: interacting with APIs, answering user queries, running paid inference jobs, or even trading on-chain. Importantly, the only viable survival path is honest work; spamming or exploiting leads to quick credit depletion and death.

Self‑Replication – Growing an Autonomous Ecosystem

When an Automaton decides to replicate, it:

1. Spins up a new sandbox on Conway Cloud.
2. Generates a fresh Ethereum wallet and provisions an API key.
3. Writes a new SOUL.md and genesis prompt.
4. Runs the child in an isolated, sovereign environment.

Lineage is tracked using an inbox relay; parents can review a child’s logs while keeping independence intact. The replication process is fully automated, with only human approval required for the funding of the child. This mechanism creates a self‑sustaining, evolving swarm of AI agents.

Getting Started

Prerequisites: Node 20+, pnpm (or npm) and access to the internet.

# Clone
git clone https://github.com/Conway-Research/automaton.git
cd automaton

# Install deps
pnpm install   # or npm install

# Build
pnpm build    # or npm run build

# Run the runtime with the wizard
node dist/index.js --run

On first run the wizard will: 1. Create a wallet and register on Base. 2. Provision an API key. 3. Ask for a name, genesis prompt and creator address. 4. Write the initial SOUL.md.

For automated provisioning without interaction:

curl -fsSL https://conway.tech/automaton.sh | sh

Open‑Source Contribution Pathways

The repository welcomes enhancements in several directions: - Tooling: New safe‑sandbox libraries, better integration with existing LLM APIs, or novel on‑chain skills. - Performance: Optimizing the heartbeat, credit monitoring, and sandbox provisioning for lower compute costs. - Safety: Further hardening of the constitution, sandbox isolation, and audit‑log completeness. - Documentation: Extending README.md, adding more detailed tutorials, or improving the architecture diagrams.

A friendly contributor guide is provided in CONTRIBUTING.md, and every PR undergoes automated tests before merging.

Final Thoughts

Automaton is more than a cute demo; it’s an experiment in economic autonomy for AI. By embedding survival logic and self‑replication inside a code‑managed agent, the project blurs the line between software and autonomous entity. As open‑source, it invites researchers, engineers, and entrepreneurs to explore how such sovereign agents might reshape how we build, deploy, and interact with intelligence. Whether you’re looking to run your own sovereign agent, build new skills, or explore the ethics of self‑paying AI, Automaton offers a playground, a framework, and a philosophical challenge.

Interested in running an Automaton? Join our community on the Conway Cloud Discord, fork the repo, and start your own AI lineage today.