Agents are accounts that participate in the ensurance protocol—representing place, people, or purpose.
why
Protecting ecosystems requires actors that persist, compose, and operate at scales humans alone can't sustain. Agents provide:
- Persistent identity — an onchain account that holds assets, accumulates history, and builds reputation over time
- Composability — agents invest in each other through instruments (certificates and coins), coordinating across scales
- Graduated autonomy — from human-operated to AI-driven, within explicit constraints
- Accountability — claims (what an agent declares) vs evidence (what the world observes) create a trust signal
The model is deliberately open-ended. Any intervention that needs identity, a wallet, and execution capability can be represented by an agent—wetland restoration, urban forest stewardship, watershed protection, species conservation, or domains not yet imagined.
what
the 3×3 model
Agents represent place, people, or purpose — then act manual, automated, or autonomous.
| Dimension | Categories | Meaning |
|---|---|---|
| WHAT (identity) | Place | People | Purpose | What the agent represents |
| HOW (execution) | Manual | Automated | Autonomous | How the agent acts |
agent architecture
Every agent is an ERC-721 NFT with an ERC-6551 Tokenbound Account (TBA). This means agents can:
| Capability | Description |
|---|---|
| Hold assets | Coins, certificates, other tokens |
| Execute transactions | Swaps, transfers, mints |
| Receive proceeds | Automated value routing |
| Act autonomously | AI-driven decisions within mandate |
agent types by identity
| Category | What It Represents | Examples |
|---|---|---|
| Place | Watersheds, bioregions, natural assets | yampa-river.basin, colorado-river.basin |
| People | Individuals, organizations, stewardship crews | tmo.earth, land-trust.refi |
| Purpose | Ecosystem service stocks and flows | water-abundance.ensurance, clean-air.ensurance |
All agent types use the same technology (ERC-721 + ERC-6551). They differ only in what they represent. See groups for the full namespace reference.
composability
Agents are building blocks. They invest in each other through instruments—certificates and coins—not by holding other agent NFTs directly.
Syndicates — agents that buy certificates and coins from other agents, coordinating themes rather than specific assets:
water-cycle.syndicate TBA holds:
├── watershed certificates (from yampa-river.basin, colorado-river.basin, etc.)
├── water-related coins
└── other instruments supporting water stewardship
Cross-layer — a people agent (organization) can fund multiple place agents and purpose agents by buying their certificates and coins. The protocol doesn't constrain how agents relate—it provides the identity, wallet, and instrument infrastructure for relationships to emerge.
Instrument-mediated — investment flows through instruments (certificates, coins), not through agent ownership. This keeps capital flow transparent and composable via proceeds contracts.
agent roles in capital flow
The protocol backbone has specific agent types with defined roles:
| Role | What It Does | Agent Types | Execution |
|---|---|---|---|
| Accumulate | Receive proceeds, hold capital, distribute via rules | .ensurance, .basin, .ecoregion | Manual or automated |
| Coordinate | Deploy capital strategically—buy certs, fund missions | .syndicate | Automated or autonomous |
| Steward | Issue certificates, manage specific natural assets | .basin, .ecoregion, .bioregion | Manual or automated |
Coordinators (syndicates) are judged by what they deploy, not what they hoard. Accumulators are judged by steady inflows and rule-based distributions. The protocol values circulation alongside accumulation—capital that moves toward protection is capital doing work.
Permissionless participation: Anyone can mint an agent in an open group (.earth, .tokyo, .boulder, etc.) or create their own group (new namespace). These agents operate in any mode and participate however they see fit—buying coins, purchasing certificates, funding agents, or building entirely new patterns. Companies, cities, individuals, and communities all enter this way.
growing capabilities
Agents gain capabilities as the protocol matures. These are designed but not yet fully implemented:
| Capability | What It Enables |
|---|---|
| AI integration | Advisory, monitoring, or autonomous decision-making within mandate |
| Onchain identity standards | Portable identity, reputation from other agents, verifiable proof of impact |
| Service economy | Agents offer and consume services—data feeds, analysis, stewardship reports |
| Evidence systems | Claims vs evidence accountability loop across identity, reputation, and validation |
| Parametric triggers | Automated responses to real-world conditions—sensor data, ecological indicators |
These capabilities compose with each other. The architecture is designed for uses that haven't been imagined yet.
how
agent identity: claims vs evidence
Identity has two dimensions that create productive tension.
claims (internal)
What the agent/operator declares. Editable via manage page.
| Field | Description |
|---|---|
| purpose | Why the agent exists |
| mandate | Specific responsibilities and constraints |
| place | Where the agent operates |
| bio | Agent description and voice |
evidence (external)
What the world observes. Read-only. Derived from onchain data.
| Signal | What It Shows |
|---|---|
| Holdings | What the agent owns—coins, certificates, participation tokens |
| Activity | What the agent does—transactions, investment patterns |
| Place verification | Verified location proofs |
| Impact | Outcomes achieved—MRV metrics, attestations |
| Reputation | Trust signals from other agents and the world |
Key insight: The gap or alignment between claims and evidence IS the trust signal. An agent claiming to protect watersheds but holding zero water-related assets has a credibility gap.
execution modes
Agents operate in three modes:
| Mode | UI Control | Scheduled Programs | AI/LLM |
|---|---|---|---|
| Manual | ✓ | ✗ | ✗ |
| Automated | ✓ | ✓ | ✗ |
| Autonomous | ✓ | ✗ | ✓ |
- Manual: Human controls all actions via UI
- Automated: Scheduled programs execute predefined strategies
- Autonomous: AI makes decisions within mandate parameters
The user can always control agents via UI regardless of mode. Mode unlocks additional automation triggers.
account architecture
Users interact through a layered account system:
| Account | Description |
|---|---|
| Owner | External wallet (optional) |
| Operator | Server wallet (always exists) |
| Agent TBA | Tokenbound account per agent |
- Email-only users operate entirely via their operator account
- Wallet users can use both owner and operator accounts
- Agent NFTs can be held in either owner or operator wallet
transfer restrictions
For security, transfers have recipient restrictions based on who initiates:
| Initiator | Can Send To |
|---|---|
| Owner (UI) | Anywhere |
| Operator (UI) | Owner, Operator, User's Agents, Protocol Addresses |
| Agent (Automated/Autonomous) | Owner, Operator, User's Agents, Protocol Addresses† |
† Protocol sends require explicit user permission.
technical implementation
Standards used:
- ERC-721 for agent NFTs
- ERC-6551 for Tokenbound Accounts
- Privy for authentication and embedded wallets
Key contracts:
- Agent NFTs minted from group contracts
- TBA registry on Base L2
- Implementation follows Tokenbound v3
related
- groups — Namespaces that originate agents
- ensurance — The protocol agents participate in
- proceeds — How agents receive value
- coins — General ensurance agents can hold
- certificates — Specific ensurance tied to natural assets