Building innovative enterprise agentic AI systems on Vultr at the Agent Arena Hackathon
Vultr’s composable cloud platform is strengthened by our ecosystem of AI builders and developers. It was on full display recently when we teamed up with Cerebral Valley for the Agent Arena Hackathon in San Francisco.
Successful agentic AI relies on two pillars: the agents themselves, and the infrastructure supporting them. Agents have progressed to a point that requires better infrastructure.
That was the focus of the Agent Arena Hackathon. Using tools like Vultr Serverless Inference, Vultr Cloud Compute, and VM backends, competitors applied Vultr infrastructure to build production-ready enterprise agentic systems.
Here are the three winning projects:
Nusa: An agent that rewrites inference kernels for Vultr VX1™ Cloud Compute
Nusa is a project built to make AI models run faster on the CPU servers that a user already has. Once the user picks a model, Nusa rewrites its slowest kernel to run on a user’s exact CPU, then returns the faster version of that model.
Nusa is an LLM running on Vultr Serverless Inference. In the hackathon, Nusa tailored models for Vultr VX1 Cloud Compute, reading a profile of the model and rewriting the kernel that took the most time. It’s built to run unattended, with every attempt running in a throwaway microVM with no network, and it resulted in up to twice-as-fast decode time on AMD’s Llama 3.1 8B. Read more about Nusa, from Alazar Shenkute, here.
Skeleton Key: Fast, cheap browser automation for sites with no API
Skeleton Key gives agents a way to access the websites you use every day – even the ones without an API. Vultr is the entire control plane, with Vultr Cloud Compute VMs, Vultr VPC, Vultr Serverless Inference, Vultr Container Registry, Vultr Object Storage, and more.
If a user wants their agent (such as Claude Code or Codex) to access their Luma account, it can’t do so, as Luma lacks an API and MCP. So Skeleton Key turns the website into an API the agent can call: It walks the site, clicks by numbered element IDs, records the site’s requests, turns each request into an operation, and runs and verifies the operations in a sandbox. Results can be published as an MCP, REST, OpenAPI spec, and downloadable Python; personal data is redacted.
Learn more about Skeleton Key, built by Carl Okpala and Jiyun Park, here.
RL Forge: Turn any paper into a reinforcement learning environment
With RL Forge, users can upload a paper or specification, and the system builds exercises with a standalone evaluation runtime for the agent to learn through trial and error.
RL Forge uses document extraction, task generation, sandbox execution, and reward verification to create an environment where agents learn by doing. It improved accuracy from 2/20 to 5/20 on a mathematics evaluation and improved successful engineering repairs from 3/20 to 7/20 on an eight-turn LEGO bridge environment. Learn more about RL Forge, by Vishnu Vennelakanti, here.
These projects all reinforced the great potential of agentic AI – provided it’s built on infrastructure that can support it. Vultr provides that critical layer, and we were excited to demonstrate it with the innovative developers and builders who will power the agentic enterprise.
