Agent Arena Hackathon SF
Projects built on Vultr at the Agent Arena Hackathon SF.
Projects built on Vultr at the Agent Arena Hackathon SF.
Projects built on Vultr at the Agent Arena Hackathon SF.

We made the world's fastest, cheapest, and most accurate browser use on Vultr and it doesn't even use vision.

Nusa: an AI agent that rewrites AMD's CPU inference kernels to make models run faster.

An always-on agent working to win your next RFP. Working on your behalf to find the right opportunities and turn your experience into a compelling response. You review, make it yours, and submit.

Ligis is an on-chain trust gate that stops an AI agent's payment before it sends money to a prompt-injection attacker — verdict in under 700ms, powered by Vultr Serverless Inference.

One Slack channel for events, office snacks, swag, and budgets—coordinated by four agents with shared memory on Vultr.

GhostRange lets AI agents investigate security incidents inside disposable Vultr cyber twins, run controlled experiments, verify fixes, and produce evidence without giving the model direct control of production infrastructure.

Type one sentence and Replay turns videos and rules from the open web into a physics-checked skill for the SO-101 robot arm.

Upgrade chamber is automated dependency upgrade testing and repair service.

Problem statement one - Safe agent execution on Vultr. A web-based agent that does real work (writes and runs code) where every action happens inside a throwaway, hardened sandbox on a Vultr VM. The control plane plans the task with Vultr Serverless Inference, dispatches each action to an isolated container, streams the evidence live to the browser, has an independent verifier judge the result from that evidence only, and then destroys the sandbox.


GeoScope uses sandboxed AI agents on Vultr to design local facilities, check land fit, and compare walking access before and after.

RLForge takes any arbitrary academic paper or spec from a verifiable field - and automatically creates an RL environment.