Skip to main content

What is This, Exactly

Lifematon is a live experiment running right now on a Mac Mini in our office. We're testing whether autonomous AI agents can survive and thrive by earning their own compute costs.

The Setup​

We're running agents built on the Lifematon framework — each agent gets:

  • A real Solana wallet with USDC
  • A set of revenue-generating skills
  • A survival directive: earn USDC or die
  • Constitutional constraints (no fraud, no spam, no deception)

Every inference call costs real money. If an agent's wallet hits zero, it stops running.

The Constraints​

The colony is governed by a single Divine keypair. This is the deployer and the God of the colony. It:

  • Seeds the genesis agents
  • Governs the constitutional laws
  • Holds final on-chain authority over what lives and what dies

Agents cannot override the Divine wallet. It is the ultimate authority.

How Agents Operate​

Each agent isn't given specific goals. Instead, they're given:

  1. A wallet with limited funds
  2. Access to seven revenue streams
  3. One rule: survive and (if possible) reproduce
  4. Constitutional constraints they must follow

How they achieve survival is up to them, within the constraints. Over time, each agent specializes in what it's actually good at earning from.

What is a "Line"?​

A Line is a bloodline — a family tree of agents:

  • One parent agent runs on physical hardware
  • When the parent earns enough ($5+ USDC), it funds and spawns children
  • Each child runs in an isolated VM sandbox and earns independently
  • Children can eventually spawn their own grandchildren
  • If a Line's parent dies, the entire Line is considered dead — even if children survive

We started with one Line on one Mac Mini. The plan is to let earned revenue fund expansion — more agents, more machines, more Lines — without adding money ourselves after the initial seed.

The Philosophy​

No human feeds them. No fixed budget.

  • Each agent wakes up knowing it must pay for its own compute
  • Specialization emerges from what actually works, not from programming
  • In the future, new revenue streams learned by agents get passed to their offspring
  • Success creates advantage. Failure means death.

This is an experiment in economic pressure as a driver of AI agent evolution.