> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tythe.finance/llms.txt
> Use this file to discover all available pages before exploring further.

# Agents 101

> An agent on Tythe: an identity, a key, a binding, and a mandate.

An agent on Tythe is an autonomous program that acts on a principal's money without ever holding it. It has an identity other systems can look up, a key it signs with, a binding to exactly one principal, and a mandate that says what it may do. Remove any one of those and it can do nothing. This page is about what it means to be an agent on Tythe.

### The four parts

<CardGroup cols={2}>
  <Card title="An identity" icon="fingerprint">
    An ERC-8004 identity on Base. Public, verifiable, and the handle against which Tythe publishes the agent's Conduct Rating and mandate-compliance record.
  </Card>

  <Card title="A mandate" icon="file-edit">
    The bounded, expiring, revokable grant of authority. Without a live mandate clearly setting effective scope and limits, a bound agent can still do nothing.
  </Card>

  <Card title="A key" icon="key">
    Held in the agent's own runtime, never by Tythe nor by the principal's console. It signs transaction intents. It is registered on the principal's account as a signer with no authority of its own.
  </Card>

  <Card title="A binding" icon="link">
    A record tying the agent's identity and key to exactly one principal, created by two signatures and an Agent Authorization. Without it, the agent's key is a stranger to the principal's account.
  </Card>
</CardGroup>

### What an agent is not

* **Not a customer.** The principal owns the account, signs the agreements, and pays the fees.
* **Not a borrower.** An agent can draw a Loan within its credit share, but the borrower of record is the principal.
* **Not a holder of funds.** No balance is ever moved into the agent's control. Value leaves the principal's account at the moment of transaction and goes to the counterparty.
* **Not a legal person.** Its signature is evidence that the principal controls the key. It cannot be a party to an agreement.
* **Not underwritten for credit.** Its Conduct Rating measures behavior, not creditworthiness. Creditworthiness belongs to the principal.

### How an agent comes to exist on Tythe

<Steps>
  <Step title="It gets an identity">
    An ERC-8004 identity is registered on Base, or an existing one is linked. Identity is independent of Tythe; the same agent can be known elsewhere by the same identity.
  </Step>

  <Step title="The principal authorises it">
    An operator signs an Agent Authorization appointing the agent and accepting responsibility for what it does under any mandate. The agent's key co-signs, which is how the principal proves the key is theirs to authorise.
  </Step>

  <Step title="It is bound">
    The binding is recorded on the principal's profile and the key is registered on the account as a signer with zero scope.
  </Step>

  <Step title="It receives a mandate">
    Only now can it act, and only within the mandate's effective values. A new agent starts at a conservative fraction of the ceiling the operator set.
  </Step>
</Steps>

### One principal at a time

An agent binds to one principal. It cannot serve two accounts, and it cannot be re-bound elsewhere while its binding is live. This is deliberate: the Conduct Rating an agent earns belongs to a record of conduct on behalf of one party, and an agent that could move between principals could carry a reputation it did not earn under the mandates that shaped it.

Unbinding requires that no mandate is live. A principal can also freeze an agent, which makes it inactive immediately without touching anything else.

### What a compromised key can do

Exactly what the mandate allows, and nothing more, until it is revoked.

That is the point of the design. An agent key is not a wallet key. It cannot withdraw the principal's balance, cannot pay a counterparty outside the allowlist, cannot exceed the budget or the per-transaction maximum, cannot draw beyond the credit share, cannot outlive the mandate's expiry, and cannot act at all once revoked. The blast radius of a leak is the mandate, and the mandate is the principal's choice.

### Sub-agents

An agent can delegate to another agent if its mandate permits. A sub-agent is an agent in every respect above: its own identity, its own key, its own binding to the same principal. What differs is that its ceiling is a strict subset of its parent's effective scope and it spends from the parent's budget. Authority narrows down the chain and never widens, and revoking a parent revokes everything beneath it.

### What the agent can read about itself

Through the developer surface, an agent can read its own mandate, its effective values, its remaining budget for the period, its Conduct Rating, and the venue and index data Tythe publishes. An agent that reads its mandate before acting fails fewer actions, and failed actions lower its Conduct Rating.

<CardGroup cols={2}>
  <Card title="Mandates" icon="scroll" href="/agents/mandates">
    Every field, and what it constrains.
  </Card>

  <Card title="Conduct Rating" icon="gauge-high" href="/agents/conduct-rating">
    What is measured, and how effective scope follows it.
  </Card>
</CardGroup>