alpha v0.0.1

a human api for agents

Build and ship production agents with a second thought. Call ask() and a real human sanity-checks your agent's work.

See how it works

One SDK · Humans included · Works with any agent framework

Works with your stack

LangChainCrewAIAutoGenOpenAIAnthropicHugging Face

How it works

One function call. That's it.

Your agent calls await ask(). Looper finds a human, gets the answer, and your agent keeps going.

With Looper

13 lines

1import { ask } from "looper"23async function handleRefund(order) {4  if (order.amount > 500) {5    // One line. Looper handles everything.6    const ok = await ask("Approve this refund?")78    if (!ok) return { status: "denied" }9  }1011  await payments.refund(order.id)12  return { status: "refunded" }13}

Without Looper

24 lines

1async function handleRefund(order) {2  if (order.amount > 500) {3    // Save state to database4    await db.savePendingApproval(order.id)5    // Build a UI for the reviewer6    await sendApprovalEmail(order)7    // Set up a webhook to catch the response8    await webhooks.register(order.id)9    // Poll or wait... somehow10    return { status: "pending" }11  }1213  await payments.refund(order.id)14  return { status: "refunded" }15}1617// Plus you need to build:18// - Approval UI for reviewers19// - State machine for pending/approved/denied20// - Webhook handler to resume the flow21// - Timeout + retry logic22// - Context restoration after resume23// - An actual pool of reviewers24// ... and maintain all of it forever

Capabilities

Everything agents need to ask.

One function call. Looper provides the humans, handles the logistics, and returns the answer.

Durable pause & resume.

Your agent calls ask() and suspends. Looper handles the rest -- finding a human, getting the answer, and waking your agent back up exactly where it stopped.

Humans on demand.

You don't recruit reviewers. Looper provides trained humans who handle your agent's questions -- approvals, classifications, content checks, whatever you need.

Built-in retries & SLAs.

Responses come back fast. If they don't, Looper retries, reassigns, or returns a configurable fallback. Set your SLA per ask() call.

Branch on human input.

Route your agent's execution based on the response. Approve, deny, escalate -- each path is just code, not config.

Audit trail by default.

Every ask(), every response, every decision is logged with timestamps and full context. Compliance-ready out of the box.

Framework agnostic.

Works with LangChain, CrewAI, AutoGen, or plain Python. Looper is infrastructure -- it plugs into whatever you're already building.

Observability

Trace every decision, human and machine.

See exactly when your agent paused, how fast a human responded, and what happened next. Every ask() is a traceable event in your execution timeline.

COMPLETE

refund-agent / handle_refund

Total

48s

Run ID

run_9xK3nR

agent.start()

4ms

db.get_order()

22ms

ask()

awaiting human

---

human.responded

47s

agent.resume()

3ms

payments.refund()

189ms

Infrastructure

Built for production agents.

You write the agent. Looper provides the humans and the infrastructure to connect them.

Durable execution.

Agent state is persisted automatically. Restarts, deploys, even crashes -- your agent picks up where it left off.

Sub-minute responses.

Looper's human workforce is always on. Most questions get answered in under 60 seconds -- no waiting on your own team.

Humans included.

You don't hire, train, or manage reviewers. Looper provides on-demand human intelligence matched to your task type.

Webhooks & callbacks.

Get notified on every event -- ask sent, human responded, timeout, fallback. Pipe results into any system.

Give your agents a second thought.

Install the SDK, add one ask() call, and Looper's humans handle the rest.

Get started Talk to sales