Ramestta AI Agent OS

Guarded mainnet betaUse controlled test allocations

Ramestta AI Agent OS is the controlled execution layer for onchain agents. It gives an agent a durable identity, a purpose-built smart wallet and explicit operating boundaries, rather than giving an LLM an unrestricted private key.

ℹ️Control remains with the owner

An agent can prepare or execute only the actions allowed by its controller policy. Owners can require approval, set value limits, pause activity and revoke permissions at any time.

Agent operating model

◈

.rama identity

A human-readable identity for discovery, attribution and communication.

◉

Smart wallet

A dedicated execution account with policy-aware permissions and transaction history.

✓

Permission layer

Scope contracts, recipients, methods, value ceilings and approval requirements.

◷

Scheduler and relayer

Run approved recurring jobs and submit authorized transactions without exposing an owner key.

Lifecycle

Create or assign a .rama identity to the agent.
Deploy or configure its controlled smart wallet.
Define allowed contracts, actions, spend limits and approval rules.
Fund a small test allocation and monitor activity.
Schedule approved tasks or send approved actions through a relayer.
Pause, update or revoke the policy when the operating scope changes.

What agents can coordinate

Capability	Purpose	Control
Onchain execution	Call allowed contracts and methods	Allowlist, value limits and approvals
Treasury operations	Spend or rebalance a defined allocation	Per-action and rolling limits
Scheduled actions	Execute recurring approved workflows	Scheduler policy and owner pause
Messaging	Coordinate through MumbleChat-compatible encrypted channels	Identity-bound communication
Audit trail	Provide a verifiable record of agent actions	Wallet and onchain transaction history

Developer entry point

Agent policy shapetypescript

const policy = {
  agentName: 'treasury.rama',
  allowedTargets: ['0xAllowedContract'],
  allowedMethods: ['0xa9059cbb'], // transfer(address,uint256)
  maxValuePerAction: '1000000000000000000',
  approvalRequiredAbove: '250000000000000000',
  emergencyPause: true,
};

Use the Agent OS control surface for guarded testing and start with a minimal policy. Do not grant broad permissions or large allocations to an unaudited workflow.