Inspiration
Hackathons reward speed, creativity, technical depth, and a polished presentation—but delivering all four under intense time pressure is difficult. We built 0xClaw to explore what happens when AI moves beyond being a coding assistant and becomes an autonomous hackathon teammate capable of managing the entire journey from idea to submission.
What it does
0xClaw is an autonomous AI hackathon competitor. Give it a hackathon URL and optional sponsor documentation, APIs, SDKs, repositories, or technical resources. It researches the event, extracts judging criteria and sponsor requirements, generates and scores project ideas, selects the strongest concept, designs the architecture, writes and tests the code, produces documentation and presentation assets, verifies submission readiness, and prepares the final submission package. Its mission-control dashboard provides live execution stages, agent activity, terminal logs, decision summaries, token and cost metrics, generated files, retries, checkpoints, and overall progress.
How we built it
We built the operator console with Next.js, React, strict TypeScript, Framer Motion, and a custom responsive design system. The interface uses an orchestration model organized around nine stages: Research → Ideate → Plan → Architect → Implement → Test → Document → Verify → Submit. Specialized agents handle research, sponsor analysis, architecture, frontend development, backend development, smart contracts, testing, security, DevOps, documentation, presentations, and submission preparation. A central orchestrator coordinates dependencies, parallel work, retries, checkpoints, and execution state. GPT-5.6 provides high-level reasoning and planning, while Codex is responsible for implementation, debugging, testing, and repository-level engineering.
Challenges we ran into
The largest challenge was designing a system that could remain understandable while many autonomous agents were operating simultaneously. We needed to expose useful progress and decisions without revealing private chain-of-thought or overwhelming users with raw internal data. Long-running execution introduced additional challenges around retries, task dependencies, crash recovery, persistent state, streaming logs, safe code execution, and cost control. We also had to design the interface so that complex workflow information remained readable across desktop and mobile screens.
Accomplishments that we're proud of
We are especially proud of creating a cohesive mission-control experience rather than another chat interface. Users can understand what the system is doing, which agents are active, what files are being generated, how much the run costs, and why major technical decisions were made. We also created a modular execution model that supports parallel agents, resumable workflows, observable progress, configurable safeguards, and clear verification gates before submission.
What we learned
We learned that autonomous software engineering depends as much on orchestration and observability as it does on model capability. Strong agents still need clear responsibilities, constrained tools, dependency management, checkpoints, verification loops, and recovery strategies. We also learned that trust requires transparency. Users do not need hidden reasoning, but they do need concise decision summaries, visible progress, predictable controls, and evidence that generated work has been tested and verified.
What's next for 0xClaw
To develop built-in support for 10 messaging platforms (Telegram and WhatsApp have now passed solid tests). Enabling any via Settings.
Log in or sign up for Devpost to join the conversation.