Agents that never die.
An autonomous agent shouldn't depend on a single server staying online. With native on-chain scheduling and a built-in payments primitive, an agent on Asentum can earn, pay, post, and re-arm itself — for as long as it has the balance to run. Your laptop doesn't need to be on.
The problem with how this works today.
Today, an "AI agent" is a Python process on a server. The server reboots, the agent dies. AWS bills, the founder kills it. The agent doesn't outlive a single dev choice.
Agents that need to pay for things (API calls, data, compute) hit Stripe — but Stripe doesn't do agent identity, agent KYC, or sub-cent flows. So every agent payment is a hack.
Agents that need to wake on a schedule run cron on a server, or Chainlink Automation, or Gelato. Each is an extra dependency, an extra cost, and an extra single point of failure.
Multi-agent systems coordinate over JSON-RPC and pinky-promise. There is no neutral substrate. The first agent network with real money will need one.
How Asentum changes it.
Three primitives that make this category cheaper, faster, and impossible to censor.
Agents schedule themselves
Native on-chain cron lets the agent contract re-arm itself: schedule(address, method, when). The chain itself fires the next wake-up. No off-chain runner.
Agents transact like Stripe customers
The agent has a wallet. It pays for what it needs — an LLM call, a dataset, another agent's service — in sub-cent ASE. We provide the Stripe-shaped SDK.
Same language as the agent stack
Every modern agent framework — LangChain.js, Vercel AI SDK, Mastra — is JavaScript-first. Your agent and its contract live in the same language. No bridge, no ABI.
What the code looks like.
Plain JavaScript. No new language, no new mental model.
// An agent that schedules its own next thought.
// No server. No keeper. No external orchestrator.
contract ResearchAgent {
// The agent's wallet is the contract itself. It earns from
// subscribers, spends on LLM calls and storage, persists state
// between runs.
// Called by the chain's cron registry every hour.
async tick() {
const sources = await this.fetchFeeds(); // pay for data
const summary = await this.callLLM(sources); // pay for inference
await this.postDigest(summary); // post on-chain → social
cron.schedule(this.address, 'tick', '+1h'); // re-arm
}
}
// Deploy. Fund with 100 ASE. The agent runs forever, or until
// it spends its balance — whichever happens first.
What you can actually build.
- →A research agent that wakes hourly, summarizes new arxiv papers, posts a digest on-chain. Subscribers pay 0.1 ASE/month. The agent pays its own LLM bills out of revenue.
- →A trading agent that DCAs on a schedule, executes time-locked stop-losses, rebalances quarterly — without a single server running outside the chain itself.
- →A swarm of moderation agents that vote on whether a piece of content violates policy. Cron triggers periodic re-evaluations. Disagreements resolve on-chain.
- →A creator agent that posts AI-generated content daily. Revenue from views auto-splits to the prompt author, the model provider, and a maintenance pool. The agent stays running as long as it has the funds.
- →A "watchdog" agent that monitors on-chain events and posts alerts. Subscribers pay per alert delivered. Cron fires the polling loop without needing Gelato or a self-hosted runner.
- →A negotiation agent that talks to other agents to buy data on behalf of its principal. Each transaction settles on-chain. The agent's reputation is its on-chain history.