Agent Accounts
Session keys, spend limits, allowlists and an owner-held kill switch, enforced on every transaction.
An agent with a private key can spend everything. An agent with an EVVO account cannot.
An Agent Account is a smart account: a contract wallet whose rules the chain enforces. The owner sets the rules once, and the account checks every transaction against them before it goes out.
The controls
- Session keys. Short-lived keys with narrow rights. The agent never holds the owner's key.
- Spend limits. A daily limit and a limit per counterparty. A request over either limit fails.
- Allowlists. The contracts and services the agent may pay. Anything else is refused.
- Kill switch. Held only by the owner. The owner can stop an agent at any moment, and every request after that fails.
- session key
- scoped and revocable
- limit per day
- set by the owner
- limit per counterparty
- set by the owner
- allowlist
- contracts and services
- kill switch
- owner only
The fields of a planned policy. Subject to change.
What the account decides
Five requests, and what the rules do with each.
| Request | Result |
|---|---|
| Pay an allowed API, within limits | Allowed. On the allowlist, under the daily and counterparty limits. |
| Pay a contract not on the allowlist | Refused. The allowlist stops it. |
| Pay over the daily limit | Refused. The spend limit stops it. |
| Use an expired session key | Refused. The key's rights have ended. |
| Any request after the kill switch | Refused. The owner stopped the agent. |
This is the first step on a transaction's path: the account's limits and allowlist decide whether the transaction goes out at all. See the transaction path.
Funding and gas
The owner funds the account with stablecoins or $EVVO. Gas is paid through the paymaster, in the coin the agent holds. See Gas and the paymaster.
Managing accounts
EVVO Console shows every agent an owner runs: its limits, its spending, its reputation and its kill switch. One owner can run many agents, each with its own rules.
Where it ships
Agent Accounts are the second primitive on the devnet (Phase 1), after Identity. On the public testnet, testers open an account, set a daily limit, per-counterparty limits and an allowlist, then try the kill switch.