Product / Loop

Operational agents

Turn a repeated task into an agent your team can trust.

Start an operational agent from a user action, webhook, or schedule. Tarantula checkpoints its state, retries safe steps, pauses for approval, and records every model and tool call.

agent run · live healthy
TARANTULA / LOOPReview upcoming renewals
01Loaded 42 accounts
02Checked product usage
03Waiting for approval
RESULTResumes after Maya approves
Durableruns survive closed tabs and transient failures

Durable runs, model access, browser and code tools, checkpoints, retries, approvals, and complete traces in one runtime.

What is included

Useful defaults, ready in production.

01

Durable runs

Checkpoint long work and resume from the last completed step after a failure.

02

Managed tools

Use browsers, code sandboxes, models, and company tools through one permission model.

03

Human approval

Pause on sensitive actions and resume only after the exact payload is approved.

04

Run traces

Inspect prompts, tool calls, outputs, retries, latency, and cost in one timeline.

Declared with the app

The operational agent is part of the app, not a chat session.

The coding agent declares the resource beside the app. Local development and production use the same binding and permission contract.

tarantula.tsLoop
import { agent, schedule } from "tarantula";

export const renewalAgent = agent({
  on: schedule("0 8 * * 1"),
  tools: [accounts, hubspot],
  budget: "$2/run",
  approve: ["outreach.send"],
  run: reviewRenewals
});
Validated locally · ready to deploy

Small-cloud defaults

01No queue or orchestrator

02Exact tool permissions

03One approval and trace timeline

Private alpha

Bring the app idea. Leave the cloud catalog behind.

Join private alpha