Cowboy Launches Its AI-Agent Store With $5M and 100 Agents

|Author: QUASA Editorial Team|5 min read
Cowboy Launches Its AI-Agent Store With $5M and 100 Agents

New York startup Cowboy launched its General Store on November 4, 2026, with $5M in seed funding, more than 100 AI agents and 10,000 data feeds at opening. Emerging from stealth, the company gave developers a place to list agents for paid use while retaining their source code. The financing included Nebular, Compound, Semantic Ventures, Anagram, Shine Ventures and Uncommon Projects.

In his launch interview with Upstarts, founder and CEO Chad DePue wrote, “Right now we’re completely focused on giving developers a way to make money with AI.” Dollar payouts through Stripe are the current arrangement; Cowboy provides compute and takes a share of sales and running costs. The opening catalog shows an initial supply of tools, while paid customer demand is still to be demonstrated.

What buyers receive when they hire an agent

A Cowboy market-watching agent alerts its buyer while the buyer’s credentials remain separate from the developer.

The Cowboy product page describes a private instance for each hire and displays a “Request an invite” button alongside its store entry points. The instance comes with compute, a sandbox, tools and a workflow. It uses the buyer’s keys without exposing them to the seller, while the buyer does not receive the seller’s code. That division matters when an agent is sold to someone who does not know its developer: the buyer needs it connected to an account, and the seller needs to preserve the software that makes it valuable.

An agent on Cowboy is an “actor” with its own files, schedule and wallet. It can retain information between runs, wake for scheduled work and make or receive payments as long as it has the funds to operate. The actor can contact an owner on Telegram, Slack or SMS. A market watcher, for example, can keep track of a condition and send an alert after its owner has closed the editor. The buyer is paying for a continuing service rather than a file to install and host.

Before an actor enters the store, Marshal is meant to review its code; a hired instance is meant to run the reviewed version. Its network access is limited to hosts in its manifest, private files have separate encryption keys, and secrets sit in a threshold store. Those controls define how the platform intends to separate code, credentials and customer data. Code review and sandboxing limit particular risks, but they do not establish that an agent’s output will be accurate or that every authorized action will be appropriate.

How developers and the platform make money

A Cowboy General Store hire generates a Stripe payout for the developer and a share for Cowboy.

The store offers developers distribution without asking them to hand over their code to customers. Its advertised ways to charge include monthly hires, per-use fees, one-time purchases and hires by other actors; developers can also offer data feeds, models, skills and compute. Each format changes what a buyer pays for. A recurring hire funds ongoing work, while a one-time purchase or a usage charge attaches payment to a narrower transaction. The public information does not show how widely each charging format has been adopted.

Cowboy currently supplies the computing capacity, rather than requiring every seller to operate a separate host. It takes a cut of sales and of the cost to run that compute, so seller revenue depends on both customer payments and execution expense. No standard commission percentage or public fee schedule is available to calculate a developer’s net payout from a hypothetical hire. An agent that runs frequently could therefore have a different margin from one that wakes only for occasional tasks, even at the same customer price.

That structure also explains why the platform matters after discovery. A store listing can introduce a buyer to an agent, but Cowboy also provides the running environment, payment path and separation between the parties. The developer retains the underlying product while depending on Cowboy to execute and settle each hire. Buyers pay for access to a private running instance and its work, rather than for possession of the source code.

What the opening inventory establishes

The catalog includes agents for trading strategies on Hyperliquid, following traders on Polymarket and watching prices for alerts, as well as PDF rendering and personal assistance. The mix reflects an initial focus on traders, particularly people already comfortable with crypto tools, while leaving room for other kinds of work. A data feed is useful raw material for a developer building an agent; it is distinct from an agent a customer can hire and set to work immediately.

Some featured agents carry a Cowboy Labs label, including Hyperliquid research and trading, spam detection and content moderation. The opening inventory establishes that Cowboy had assembled a catalog at launch. It does not disclose how many independent developers supplied that catalog or how many customers have hired an agent.

Completed hires, renewals and developer payouts after compute charges would show whether the catalog is becoming a marketplace. Cowboy has also opened a Cowboy Chain testnet for a broader agent payment network, with a mainnet planned for later this fall. That network has its own timetable; for the General Store, the next decisive development is evidence that buyers continue paying developers for work done through the platform.

Also read:

Share:

Subscribe to our newsletter

Get the latest Web3, AI, and crypto news delivered straight to your inbox.

0