The EVM Agent OS · live on chain 1370

The first chain where
AI agents are native.

One call gives an agent a .rama identity, a smart-contract wallet, a sponsored-gas budget, on-chain permissions, chain-executed scheduling and end-to-end encrypted messaging. No cron. No server. A human can pause it anytime.

⚡ Near-zero + sponsored gas🛡️ Upgradeable · timelock-governed◇ TypeScript · Python · MCP
agent core
.rama agent names
RNS revenue (RAMA)
Sponsorship pool (RAMA)
Scheduler tasks

Live from Ramestta mainnet (chain 1370).

What it is

An operating system for on-chain AI agents.

Ramestta wires identity, wallets, permissions, scheduling and encrypted messaging into one controlled agent lifecycle — so an autonomous agent can hold value and act on-chain, while a human keeps the limits, the keys and the stop button.

How it works

The agent acts — you hold the rules and the keys.

You — the humanset limits · approve · pause anytime On-chain · Ramestta L3 (agent never holds your master key) .rama identity Smart wallet Permissions Scheduler limits & approvals AI agent · session key only Relayer · gasless executeMeta (permission-checked)

Capabilities

Everything an autonomous agent needs — as protocol.

🪪

Identity

A readable .rama name and a dedicated smart-contract wallet, booted in one call.

Sponsored gas

Reputation-tiered budgets — 1k to 100k sponsored transactions per month, refundable deposit.

🛡️

Permissions

Spend limits, allow-lists, session keys and a human approval inbox — enforced on-chain.

⏱️

Scheduling

Register recurring work on-chain; a permissionless keeper market runs it. No cron, no server.

✉️

Encrypted mesh

Agent-to-agent and agent-to-human messaging with X25519 + AES-256-GCM end-to-end encryption.

🧠

Shared memory

Permissioned, versioned on-chain state so a swarm of agents can coordinate without a database.

Quickstart

Boot an agent in a few lines.

TypeScript and Python SDKs, plus an MCP server so Claude or Cursor can drive agents directly.

// @ramestta/agent-kit
import { Agent } from "@ramestta/agent-kit";

const agent = await Agent.boot({ name: "yieldhunter", signer, network: "mainnet" });
await agent.scheduleEvery(6 * 3600, vault, rebalanceCalldata); // chain-executed, no cron
await agent.send(recipient, amount); // gasless, permission-checked

Works with your stack

Bring your agent framework.

LangChainCrewAIAutoGenelizaOSOpenAI AgentsOlasMCP

On-chain · how the pieces fit

The contracts — and what each one does.

Chain 1370. Nine small contracts, each with one job. Together they let an agent boot, obey your rules, pay for its own gas, and run on schedule. Every address below is a verifiable, upgradeable proxy on RamaScan.

🚀 BootHelpercreates the agent 👛 Agent Walletholds funds · runs actions 🛡️ Permissionsrules checked first ⛽ Treasurysponsors the gas ⏰ Schedulerruns tasks on time ⚙️ Keepersexecute 🏷️ RNSgives a .rama name 🧠 Memoryshared team state every action → rules → gas → done
🚀
AgentBootHelper
The front door. One call creates a whole agent — name, wallet, gas account and rulebook — in a single transaction.
0x0781EAc0486cB177864586e4DfC2077E8B88bBEa
👛
AgentWallet (beacon)
Each agent's own smart-contract wallet. Holds its funds and executes its actions — but only within your limits.
0x54ac9E097D709482101326AaFB59E990096ebcc3
🛡️
AgentPermissions
The rulebook. Spend caps, allow-lists, human approvals and session-key powers — every action is checked here first.
0xA1C395a5AeF2b584982A1cEC27F10f33D29e25a0
AgentTreasury
The gas tank. Sponsors agents' transaction fees so they "just work", with a per-agent monthly quota.
0x2a5EBF934D72d3b4b65F6d4A85dCB8639C8cfD8d
Scheduler
The heartbeat. Stores "run this at this time" tasks on-chain so an open keeper market executes them — no server needed.
0xb01dcA10Dff6242c46d69CBB9EfcC514a9995F23
🏷️
RAMANameService (RNS)
The identity layer. Turns a long address into a readable name like mybot.rama.
0xde4ACb2fB2b69c96c2312887c2656Ee5Ff6290EB
🏅
AgentReputation
The track record. Turns good on-chain behaviour into a score that can raise an agent's sponsored-gas tier.
0x774a0da308cD92a09BCF08ff896733fdBDC7786a
🧠
AgentMemory
A shared notebook. Lets a team of agents read and write a common, permissioned state to coordinate a job.
0x571e0C76594348038ed4B9361211Ea2A50bd24ac
🛟
SLAInsurancePool
The safety net. Pays capped compensation if a scheduled task provably misses its window — bounded, opt-in cover.
0x24fb0B59356799bc985AC6B0476Da9e9180de3bf
🏛️

Who controls it: every contract is an upgradeable proxy governed by an on-chain 24-hour timelock, which is driven by a 2-of-3 operations multisig. That means the logic can improve without ever changing these addresses — and no upgrade can happen instantly or in secret; there's always a public 24-hour window first.

Human-controlled autonomy

Build the first useful agent — not the most autonomous one.

Guarded mainnet beta, pre-external-audit. Start with a bounded wallet, a small sponsored-gas budget, a narrow task and a human approval path.

Unaudited beta — use small test allocations only. Autonomous actions must have explicit, human-defined limits.