Today we're launching Twin publicly, after a 1-month beta where users deployed more than 100,000 fully autonomous agents. We're also announcing a $10M seed round led by LocalGlobe.
Over the past year, tools like Cursor, Claude, and Lovable have shown that AI will soon write all the software. But most businesses don't run on code. They run on operations: selling, managing customers, handling finances, marketing, back-office, logistics. Twin is the last missing piece.
We saw during the beta users one-shoting complex workflows in minutes (instead of days or weeks), without touching a line of code. But the most incredible use cases we've seen go much further — some users built fully autonomous businesses entirely on Twin.
We saw glimpses of what the future holds. Here are a few impressive agents built by the community:
Autonomous Trading System
Exchanges, X, CoinGecko, Telegram, Email
A self-learning trading system that scans markets, decides when to act, executes trades, manages risk, tracks P&L, and continuously retrains itself — operating like a small hedge fund with no humans in the loop.
Autonomous Plumbing Business
Phone, Google Calendar, Airtable, Slack, Telegram, Email
A local service business that answers calls, classifies emergencies, books jobs, dispatches technicians, tracks work in a CRM, and keeps the owner informed — running daily operations autonomously.
Retail Arbitrage Business
Amazon, Vinted, Payments
An agent that scans Amazon for underpriced products, evaluates resale margins, automatically purchases inventory, and creates optimized listings on Vinted — turning price inefficiencies into profit without human involvement.
Autonomous Flower Import Business
Microsoft Teams, Email, Excel, Suppliers
An agent that runs a wholesale import operation by parsing orders, negotiating with farms, optimizing prices, confirming supply, coordinating deliveries, and handling invoicing end to end.
What's most surprising is that most Twin users are non-technical. They don't write code or understand APIs. What previously required teams of 10s of engineers to wire integrations, manage infrastructure, and engineer prompts can now be done by anyone with an idea, in just a few minutes. Twin writes the integrations, fixes errors, and maintains them over time, without any user intervention.
Twin agents have long-term memory that works more like a human brain than simple chat storage. An internal memory agent consolidates what matters, prunes what doesn't, and maintains context across hundreds of tasks. This memory is shared across agents — and soon across users — allowing the system to improve exponentially.
As agents learn, users spend less time prompting them, and agents become more efficient over time. Twin uses high-reasoning models like Opus when planning and building, then automatically switches to smaller models during execution, at one-tenth the cost. We've seen agents perform 150+ tasks in a single run while staying below $1. Combined with long-term memory and cost efficiency, production agents will soon be able to tackle tasks for days.
We decided to build Twin in the cloud so you can start in seconds, with no setup, and scale infinitely. Our remote machines support millions of concurrent tasks and browser sessions, and you don't have to manage security yourself. In under a minute, you can get your agent started.
Twin is now available to anyone. We'll keep posting the most interesting use cases we discover from the community.
Excited to see what you'll build!