HomeProjectsHackathonsEventsWorkBlogTimeline

Open Agents 2026

An async ethglobal hackathon focusing on Agents and Web3

Remote
ETHGlobal
April 24, 2026
Event Date
14 days
Duration
3 members
Team Size
Project Built
Experience Ratings
Food
5
Swag
5
Stay
5
Mentorship
5
Whoops. forgot to take any pics

14 days. Async. Remote. No sleep schedule, just GitHub notifications and a shared DigitalOcean droplet slowly growing in complexity.

ETHGlobal OpenAgents was a 2-week async hackathon, and honestly, that format hits different. No venue, no swag bag, no free food to keep you going at 3AM. Just you, your team, and the problem. There's something weirdly freeing about that, but also no structure means no guardrails either.

We built Crucible: a browser-based agentic development environment for Web3. The pitch: describe a dApp in natural language, and the agent writes the Solidity contracts, compiles them, deploys to a local chain, wires up a frontend, and renders a live preview, all in one browser tab. Think v0 by Vercel, but for decentralized apps. The agent has long-term memory, a peer mesh to share solutions across instances, and a post-deploy automation layer through KeeperHub's MCP integration. Basically everything a Web3 dev actually needs, without the usual juggling act of five disconnected tools.

The scale of it was kind of absurd for a hackathon. Seven custom MCP servers, a per-workspace Hardhat node running server-side, a real PTY-backed terminal (not a fake one, an actual bash shell inside a Docker container piped through xterm.js over WebSocket), a live frontend preview server per workspace, and a peer mesh via Gensyn AXL so agent instances could share revert fixes with each other in real-time. All of it had to talk together.

The deployment story was genuinely fun to figure out. We ran everything on a single DigitalOcean VPS droplet, multiple Docker containers sharing the same host, each workspace spinning its own runtime container. For the KV stores and persistent agent memory, we used AWS nodes. Getting all of it to not fall over under load, with proper routing between containers and sane port management using Portless for the local dev URLs, was a proper systems engineering problem. The kind you don't usually get to touch in a hackathon.

The team collab was entirely async, all over GitHub. PRs, reviews, task lists, architecture docs, everything. We wrote proper spec documents before touching a line of code. Fourteen days sounds long until you're building something at this scale with no shared office hours, then it starts to feel very short.

Didn't place. Didn't even make the finalist list.

That stung a little, not gonna lie. The amount of effort that went into Crucible was real, and we shipped an actual working demo with live deployment and everything. But that's how it goes with ETHGlobal sometimes. Hundreds of solid projects, a handful of prizes, and a lot of really good work that just doesn't make the cut.

The one bright spot: KeeperHub gave us a shoutout. Not a prize, just a mention, but given that our KeeperHub integration was load-bearing rather than decorative (the agent drove their MCP natively to wire on-chain automation after every deploy), it was genuinely nice to know someone noticed. Sometimes that's enough.

Still proud of what we shipped. Crucible was the most technically complex thing we'd built as a team, and it actually worked.