Nous Research's Hermes Agent Can Now Spend Real Money Autonomously
Nous Research, NVIDIA, and Stripe launch a $17,500+ hackathon to build AI agents that can autonomously earn, spend, and run real business operations using Hermes Agent's new financial and safety integrations.

- Hackathon launched: Nous Research, NVIDIA, and Stripe are running a joint hackathon for agents that can earn, spend, and run real business operations.
- Prize pool: $17,500 in cash plus three NVIDIA DGX Spark workstations and up to $9,000 in Stripe credits across top three winners.
- New Stripe Skills: Hermes Agent now supports native Stripe integration -- agents can buy services, pay APIs, and provision SaaS with configurable spend limits and human approval gates.
- NVIDIA stack integrated: NemoClaw (secure sandboxed runtime), Nemotron 3 Ultra (550B param model, 5x faster inference, 30% cheaper), and CUDA-X agent skills are all now available inside Hermes.
- Hermes momentum: The MIT-licensed framework crossed 140,000 GitHub stars in under three months and is the most-used agent on OpenRouter.
- Deadline: Submissions due end of day June 30 via X demo video, Discord, and a Typeform submission.
Hermes Agent -- the open-source agent framework that crossed 140,000 GitHub stars in under three months -- just got a lot more interesting for builders. Nous Research has kicked off a hackathon in partnership with NVIDIA and Stripe, challenging teams to build agents that can operate like a real business: earning revenue, spending money, and running operations autonomously. The timing is not accidental. It lands right as two major integrations -- NVIDIA's full agent stack and Stripe's agentic payment primitives -- become available natively inside Hermes.
The Stack Behind the Challenge
To understand why this hackathon matters, you need to understand what just got wired together. Hermes Agent is an open-source autonomous AI agent built by Nous Research that lives on your server, remembers what it learns, and gets more capable the longer it runs. It is not a wrapper around a single model API. It runs on top of 200+ models via OpenRouter, Nous Portal, NVIDIA NIM, OpenAI, or a custom endpoint.
The hackathon announcement bundles three distinct NVIDIA integrations into the Hermes ecosystem:
- NemoClaw (safety): NVIDIA NemoClaw is the open-source blueprint that ties the environment together. With a single command, NemoClaw installs the OpenShell runtime -- providing a secure environment for running autonomous agents like Hermes Agent more safely alongside open-source models like Nemotron. Think of OpenShell as a sandboxed jail for your agent: NemoClaw handles task decomposition, multi-agent delegation, tool invocation with error recovery, and policy enforcement through OpenShell -- a sandboxed runtime that prevents agents from accessing unauthorized system resources.
- Nemotron 3 Ultra (speed): NVIDIA's most capable model yet -- Nemotron 3 Ultra -- has 550 billion total and 55 billion active parameters and is the final and best model of the Nemotron 3 family. It employs a Mixture-of-Experts Hybrid Mamba-Attention architecture. In practice, it delivers up to 5x faster inference and lowers the cost of complex agentic tasks by up to 30%. Hermes users can already access it free via the Nous Portal.
- NVIDIA Agent Skills: NVIDIA is working closer to the orchestration layer where models become useful agents, with memory, tool use, routing, privacy rules and security controls. This includes NemoClaw blueprints, OpenShell secure runtime technology, and CUDA-X libraries that can be exposed as skills for agents.
The Stripe side is equally significant. In partnership with Stripe, Hermes Agent now supports a full suite of Stripe skills. Your agent can buy things, pay per-call APIs, and provision its own SaaS, with configurable safety limits on every action. This solves a problem that has quietly blocked autonomous agents from doing real work: the moment a task requires spending money -- provisioning a cloud database, upgrading a hosting tier, purchasing API credits -- most agents hit a wall. They have to stop, notify a human, wait for approval, and hope the human acts fast enough to keep the workflow moving. That bottleneck is increasingly the limiting factor for autonomous AI workflows.
The mechanism Stripe uses to solve this is scoped credentials. Stripe Projects shifts the model to support delegated agent identities -- where an agent is issued scoped credentials tied to a specific Project, complete with pre-defined permissions and spending guardrails. On the payment execution side, the Stripe Link skill wraps the link-cli so Hermes can complete purchases using one-time-use virtual cards or Shared Payment Tokens. Every spend is gated by an in-app approval in the Link mobile or web app -- Hermes cannot self-approve.
Why This Combination Is New
Individually, none of these pieces are brand new. What is new is their integration into a single, coherent agent stack. Before this, building an agent that could safely browse the web, call tools, manage memory, and spend money required stitching together five different systems with no shared security model. Now there is a reference architecture: Hermes as the harness, NemoClaw and OpenShell as the secure runtime, Nemotron 3 Ultra as the reasoning engine, and Stripe Skills as the financial layer.
Nous Research is also a member of the Nemotron Coalition -- NVIDIA's consortium of AI labs co-developing open frontier models. Nous Research is joining the Nemotron Coalition, contributing unique strengths spanning data, training environments, evaluation frameworks, and domain expertise to support the collaborative development of an open frontier model trained on NVIDIA DGX Cloud, which will serve as the foundation for the upcoming Nemotron 4 family. The hackathon is, in part, a public stress test of that partnership.
The Prize Structure
The prizes are substantial and hardware-heavy, which signals that NVIDIA is using this as a DGX Spark distribution channel as much as a community event:
| Place | Cash | Hardware | Stripe Credits |
|---|---|---|---|
| 1st | $10,000 | NVIDIA DGX Spark | $5,000 |
| 2nd | $5,000 | NVIDIA DGX Spark | $3,000 |
| 3rd | $2,500 | NVIDIA DGX Spark | $1,000 |
Every winner gets a DGX Spark -- NVIDIA's compact workstation-class AI computer -- which is exactly the hardware optimized for always-on local use, making NVIDIA RTX PCs, NVIDIA RTX PRO workstations, and NVIDIA DGX Spark the ideal hardware to run Hermes at full speed, around the clock. The Stripe credits are also meaningful: they give winners real budget to run their agent in production, not just demo it.
What Judges Want to See
Submissions are judged by all three companies on usefulness, viability, and presentation. The prompt is deliberately open-ended -- "a fully automated company or a framework to accelerate enterprise functions" -- but the Stripe integration hints at what will score well. Agents that actually transact, not just simulate transactions, will stand out. The most compelling submissions will likely demonstrate:
- An agent that autonomously provisions and pays for its own infrastructure (e.g., spins up a database, pays for the tier upgrade, monitors costs)
- A multi-agent workflow where sub-agents handle specialized tasks and settle payments between themselves
- An enterprise automation that replaces a human approval loop with scoped Stripe credentials and configurable spend limits
- A framework that other builders can drop into their own Hermes setup to add financial autonomy
How to Enter
The submission process is straightforward. Post a 1-3 minute demo video on X tagging @NousResearch with a short writeup, drop the link in the Nous Discord submissions channel, and fill out the submission form. Submissions close end of day Tuesday, June 30.
To get started with the full stack, the NVIDIA technical blog has a working end-to-end tutorial for deploying Hermes inside NemoClaw with OpenShell, including snapshot and restore scripts for persisting agent state across deployments. The Stripe Skills are available directly through the Hermes Agent docs.
The deeper story here is not the prize money. It is that three major players are converging on a shared vision of what a production-ready autonomous agent looks like: sandboxed execution, persistent memory, self-improving skills, and real financial agency. The hackathon is the first public test of whether that vision holds up when builders get their hands on it.