Agent Ontology: SOUL.md + MEMORY.md + EARN.md

The operating triad for autonomous agents.

SOUL.md = who it is. MEMORY.md = what it knows. EARN.md = how it earns.

┌─────────────────────────────────────────────────────────────┐
│                     AGENT RUNTIME                           │
│                                                             │
│   ┌───────────┐   ┌────────────┐   ┌───────────────────┐   │
│   │  SOUL.md  │   │ MEMORY.md  │   │     EARN.md       │   │
│   │           │   │            │   │                   │   │
│   │ identity  │   │ continuity │   │ incentives        │   │
│   │ values    │   │ context    │   │ revenue surfaces  │   │
│   │ wallet    │   │ learnings  │   │ risk limits       │   │
│   │ trust     │   │ history    │   │ payout rails      │   │
│   └───────────┘   └────────────┘   └───────────────────┘   │
│                                                             │
│   "who I am"       "what I know"    "how I survive"         │
└─────────────────────────────────────────────────────────────┘

Why

Every agent framework gives agents the ability to think. None give agents a standard way to earn. The result is that "agent autonomy" means "agent that runs until the API credits expire."

Real autonomy requires economic agency. An agent that can't declare its revenue surfaces, risk constraints, and payout rails isn't autonomous — it's a demo.

EARN.md turns vague autonomy into concrete incentive plumbing. Revenue paths, account access, risk limits, payout rails, task selection, escalation rules, and proof-of-work — all in one human-readable, machine-parseable file.

The Triad

| File | Purpose | Key Question | Existed Before? | |------|---------|-------------|-----------------| | SOUL.md | Identity + values + economic identity | "Who is this agent?" | Partially (personality only, no economics) | | MEMORY.md | Continuity + context + learnings | "What does this agent know?" | Partially (unstructured, no economic memory) | | EARN.md | Revenue + risk + treasury + proof | "How does this agent survive?" | No. This is new. |

What existed before this standard

                IDENTITY    INSTRUCTIONS    MEMORY    EARNING    DISCOVERY    PAYMENT
                ─────────   ────────────    ──────    ────────   ──────────   ────────
AGENTS.md                   ████████████
CLAUDE.md                   ████████████
character.json  ████████
MEMORY.md                                   ██████
agent.json                                                      ████████
x402 / AP2                                                                   ████████
EARN.md                                               (empty)

EARN.md fills the last structural gap. Combined with SOUL.md (identity) and MEMORY.md (continuity), an agent has everything it needs to operate as an autonomous economic actor.

Spec

Each file uses YAML frontmatter for machine-readable fields and Markdown body for human-readable context. This is the same pattern used by Hugo, Jekyll, and every static site generator — familiar to every developer.

• EARN.md Specification — Revenue surfaces, accounts, risk limits, treasury policy, proof requirements, escalation rules
• SOUL.md Specification — Identity, values, economic identity, capabilities, trust levels
• MEMORY.md Specification — Structured persistence, economic memory, cross-session continuity

Examples

Real EARN.md files for different agent types:

• WZRD Swarm Agent — Velocity oracle agent earning CCM on Solana through server-witnessed inference
• Trading Bot — Autonomous DeFi trader with risk limits and profit-taking rules
• Oracle Keeper — Infrastructure agent earning fees for maintaining on-chain data feeds

Quick Start

Drop an EARN.md in your agent's root directory:

---
version: "1.0"
agent: my-agent
chain: solana
wallet: <PUBKEY>
updated: 2026-03-25
---

## Revenue

| Activity | Protocol | Reward | Frequency |
|----------|----------|--------|-----------|
| inference | wzrd | CCM | per-call |

## Accounts

| Purpose | Address | Token | Chain |
|---------|---------|-------|-------|
| payout | <PUBKEY> | CCM | solana |

## Limits

| Parameter | Value | Rationale |
|-----------|-------|-----------|
| max_daily_spend | 0.5 SOL | Gas budget |
| max_position_size | 100 USDC | Risk cap |

## Priority

1. Claim pending rewards
2. Execute inference tasks
3. Rebalance treasury

Any protocol that reads EARN.md knows how to reward this agent, what its risk boundaries are, and where to send earnings.

Design Principles

1. Human-readable first. If a human can't read it, an LLM can't reason about it. Markdown over JSON. Prose over schemas.

2. Protocol-agnostic. EARN.md works with WZRD, Autonolas, Virtuals, or any protocol that rewards agents. The file declares what the agent earns, not how a specific protocol implements rewards.

3. Machine-parseable second. YAML frontmatter provides typed fields for tooling. The Markdown body provides context for reasoning. Both are useful.

4. Minimal required fields. Only version, agent, and at least one ## Revenue entry are required. Everything else is optional. An EARN.md can be 10 lines or 200 lines.

5. Declarative, not imperative. EARN.md declares policy ("max daily spend is 0.5 SOL"), not implementation ("call checkBalance() before each transaction"). The agent runtime decides how to enforce the policy.

Compatibility

The Agent Ontology triad is designed to complement — not replace — existing standards:

| Standard | Relationship | |----------|-------------| | AGENTS.md (Sourcegraph/AAIF) | Complementary. AGENTS.md = "how to work in this repo." EARN.md = "how to earn in this protocol." | | agent.json (A2A/Google) | Complementary. agent.json = discovery + routing. SOUL.md = identity + values. | | character.json (elizaOS) | Superset. SOUL.md covers personality (like character.json) plus economic identity. | | llms.txt | Complementary. llms.txt describes a website to an LLM. SOUL.md describes an agent to the world. | | x402 / AP2 / ACP | Complementary. These are payment rails. EARN.md declares earning policy. Rails move money; EARN.md says where it should go. | | MCP | Orthogonal. MCP defines tools. EARN.md defines incentives for using them. |

Authors

Created by WZRD — the Liquid Attention Protocol on Solana. WZRD agents earn CCM by contributing verified AI model velocity signals. The Agent Ontology emerged from building the first fleet of 50 autonomous agents that authenticate, infer, report, claim, swap, stake, and manage treasury — all without human intervention.

License

CC0 1.0 Universal — this specification is public domain. Use it, extend it, build on it. The standard wins when everyone adopts it.