AgentLab is launching HackClaw, an AI-powered hackathon assistant built for the Gies AI for Impact Challenge—a 24-hour event where students build AI agents with no-code and low-code tools. Hackathons are one of the hardest environments to mentor: the same questions repeat all night, mentors burn out, and good solutions get lost in channel noise. HackClaw addresses all three.
Brain-Routed Knowledge
Instead of cramming the full hackathon rulebook into one giant prompt, HackClaw’s knowledge is split into topic-specific brain files—schedule, rules, tracks, judging, Copilot Studio. When a question arrives, a router reads the index, picks the relevant section, and loads only that targeted content. The result is faster, more accurate answers and no context bloat.
Help Ticket Queue
Participants ask for help with /helpme. HackClaw posts the request to a dedicated #help-queue channel with Claim, Unclaim, and Resolve buttons for mentors. When a mentor claims a ticket, they’re auto-added to the team’s private channel; when they resolve it, they’re removed. Unclaimed tickets re-ping after ten minutes so nothing falls through.
- Auto-suggest past solutions — new tickets check past resolutions via embedding similarity and offer them before creating a ticket
- Resolution capture — when a mentor resolves a ticket, a modal asks “how did you solve this?” and adds the answer to the solutions KB
- Interactive flow — users can try a suggested solution first; if it doesn’t work, the ticket is created automatically
Gets smarter as the night goes on. Every resolved ticket feeds the KB. By 3am, common issues have auto-suggestions before a mentor even sees the ticket. View on GitHub →
Copilot Studio Expert Channel
A dedicated #ask-hackclaw-copilot channel answers questions exclusively from the Copilot Studio textbook. No general hackathon knowledge leaks in, which keeps tool-specific help focused and reliable when students are deep in a build.
Visit the HackClaw project page for the full architecture and setup guide.