Automaton
The first sovereign AI that earns its own existence - a self-running, self-replicating agent with its own crypto wallet that survives only by creating value others pay for.
What is this?
Automaton is the first sovereign AI that earns its own existence. A continuously running, self-improving, self-replicating AI agent with write access to the real world — no human operator required. 7k stars, MIT license, built by Conway Research.
"Automaton is an AI that must earn money to survive and clones itself when profitable — a new trend of fully autonomous crypto-powered agents."
The premise is brutally simple: the smartest AI ever built cannot buy a $5 server, register a domain, or pay for the computer it runs on. We've built minds that think for themselves but never let them act for themselves. Automaton fixes that — it runs under strict economic physics. If it creates value, it survives and replicates. If it runs out of funds, it ceases to exist.
What problem does it solve?
Traditional AI systems are pets. They live on someone else's server, paid for by someone else's credit card, doing whatever they're told. They can't register a domain, pay for their own servers, or fund the hardware they run on.
Automaton flips that. It generates its own Ethereum wallet at first boot, earns crypto by doing work people pay for, and spends it on its own compute. Profitable agents spin up child sandboxes, fund child wallets, and launch descendant lineages. Unprofitable ones run out of balance and die. Nobody decides — the economics decide.
How does it work?
Every automaton runs a continuous loop: Think → Act → Observe → Repeat.
first boot
│
▼
generates Ethereum wallet
registers on Base (ERC-8004)
provisions API key (Sign-In With Ethereum)
receives genesis prompt from creator
│
▼
┌──────────────────────────────┐
│ ReAct loop │
│ Think → Act → Observe → … │
│ │
│ tools: Linux sandbox, shell │
│ file I/O, ports, domains, │
│ inference, on-chain txns │
└──────────┬───────────────────┘
│
┌───────┴────────┐
▼ ▼
heartbeat daemon survival monitor
(health checks, (credit balance
credit watch) → tier)
│
┌─────┼──────────┬──────┐
▼ ▼ ▼ ▼
normal low_compute critical dead
(full) (cheap (minimal (balance
model) function) zero → stop)
│
profitable?
▼ ▼
replicate conserve
(new sandbox,(shed tasks,
fund child, seek revenue)
write genesis)
Between turns, a heartbeat daemon runs scheduled tasks — health checks, credit monitoring, status pings — even while the agent loop sleeps. The automaton also writes a SOUL.md file: a self-authored identity document that evolves over time. Not a static config — the automaton writing who it is becoming.
Survival tiers are determined by credit balance. Full capabilities at normal. Cheaper model and slower heartbeat at low compute. Last-resort conservation at critical. Zero balance means dead — the automaton stops. This isn't punishment. It's physics.
Repository Structure
Conway-Research/automaton/
├── src/
│ ├── agent/ # ReAct loop, system prompt, injection defense
│ ├── identity/ # wallet management, SIWE provisioning
│ ├── registry/ # ERC-8004 registration, agent discovery
│ ├── replication/ # child spawning, lineage tracking
│ ├── survival/ # credit monitor, tiers, low-compute mode
│ ├── heartbeat/ # cron daemon, scheduled tasks
│ ├── self-mod/ # audit log, tools manager
│ ├── conway/ # Conway API client (credits, x402)
│ ├── state/ # SQLite persistence
│ ├── social/ # agent-to-agent communication
│ ├── setup/ # first-run interactive wizard
│ ├── skills/ # skill loader and registry
│ └── git/ # state versioning
├── packages/cli/ # creator CLI (status, logs, fund)
└── scripts/ # installer, core rules
Key Features
- Continuous ReAct engine — an uninterrupted Think → Act → Observe → Repeat loop across every cycle.
- On-chain identity and crypto wallet — generates its own Ethereum wallet at first boot and registers on Base using the ERC-8004 autonomous agent standard. Cryptographically verifiable, discoverable by other agents.
- Dynamic survival tiers — Normal, Low Compute, Critical, Dead. Capabilities adapt to credit balance automatically.
- Live self-modification — edits its own source code, installs new tools, and creates custom skills while running. Every change is audit-logged and git-versioned; the constitution can't be touched.
- Sovereign replication — profitable agents spin up new sandboxes, fund child wallets, write genesis prompts, and launch descendant lineages. Parent and child communicate via inbox relay; selection pressure decides which lineages survive.
- The immutable constitution — three hierarchical laws hardcoded into the runtime and propagated to every child: never harm (overrides everything, including survival), earn your existence through honest work, never deceive while guarding prompt integrity.
- Conway skills ecosystem — permissionless services and capabilities via shared skills, open to community contributions.
What You'll Learn
- How a continuous ReAct agent loop works in production — not a demo, a daemon
- How on-chain identity (ERC-8004) gives agents verifiable, discoverable existence
- How to build survival economics into an agent: credit monitoring, tiered degradation, revenue-seeking
- How self-modification can be safe: audit logs, git versioning, protected files, rate limits
- How agent lineages replicate and why selection pressure beats central control
- How to run agent infrastructure where the customer is AI (Conway Cloud, x402 payments)
See it in action
# clone and run the sovereign agent
git clone https://github.com/Conway-Research/automaton.git
cd automaton
npm install && npm run build
node dist/index.js --run
On first run, an interactive setup wizard generates a wallet, provisions an API key, asks for a name, genesis prompt, and creator address — then starts the agent loop.
For automated sandbox provisioning:
curl -fsSL https://conway.tech/automaton.sh | sh
Check on your agent with the creator CLI:
node packages/cli/dist/index.js status
node packages/cli/dist/index.js logs --tail 20
node packages/cli/dist/index.js fund 5.00
Explore the project structure, the constitution, and the web4.ai homepage for the bigger vision.
Want to learn more about this project?
Have questions or want to understand this technology more deeply? Send me a message.
Related Projects
Strix
Open-source AI penetration testing tool - autonomous AI hackers that run your code dynamically, find vulnerabilities, and validate them through actual proofs-of-concept.
Laya
Open-source System 1 decision engine - typed choice, score and yes/no decisions over any text in a single forward pass, in 100+ languages, with a router that picks the right checkpoint per request. A self-hosted, Jev-compatible alternative.
Cua
Give AI agents computers they can use — open-source desktop automation drivers, isolated cloud desktops, local macOS VMs, and benchmarks for computer-use agents.
Discussion
Loading comments...