OmniClaw — Devpost Submission
Inspiration
Most AI assistants are reactive single-purpose tools, which honestly is no different from if I made like a basic N8N workflow. I wanted one agent that knows my whole life and proactively connects academic deadlines, market news, or other functions like note taking and weather from Discord, then chain them all together into briefings or actionable tasks like a second Brain + Hands.
What it does
Personal copilot triggered by Nanoclaw Bot in Discord. Tracks Canvas and Coursera deadlines through google calendar, monitors portfolio and AI news, drafts emails if needed, creates calendar events, sends proactive morning briefings and class reminders with zoom link if required, and tracks my projects through github mcp.
How we built it
Forked NanoClaw, added seven container skills that pertain to existing nanoclaw abilities like agent-browser and scheduled tasks (academic, market-intel, actions, project-manager, prediction-markets, cumtd-bus, weather using scraping + openweathermap API) and a central CLAUDE.md, wired three MCPs in ClaudeCode and nanoclaw (Google Workspace, GitHub, ComputeEdge), deployed to Hetzner CX33 via Baadal AI.
Challenges we ran into
Baadal's write tools threw -32602 for 30 minutes, seemed to be an API token issue but was hard to debug as there are no outputs for MCP functions. DISCORD_TOKEN vs DISCORD_BOT_TOKEN naming mismatch silently broke auto-registration I believe for a while a bit.
Accomplishments that we're proud of
Three live MCPs, seven+ composable skills, isolated containers per message, cross-domain reasoning linking AI news to held stocks and deadlines to study blocks. Deployed to production in one night. I was particularly proud that it was able to seamlessly link all the skills and MCP/skills that I gave it and suggest new interactions. For example it was able to remember I had a backlog doc for project ideas and whilst discussing new ones it would add to it. It also automatically pulled up D1 baseball news for the day/week when I mentioned I bet on UCLA winning D1 baseball on Kalshi. Things like that make it super cool.
What we learned
I learned how to setup nanoclaw! First time using any claw actually, and it was really easy to expand its features using a host of SKILL.md and MCP servers, was awesome. Nanoclaw is still very slow though, and it is hard to tell if things are working when it stops for like a minute to configure things, but when it actually works it is awesome.
What's next for OmniClaw
Mount Obsidian to repo or remote vps for more notetaking capabilities and integration into my knowledge base, open-source skills, maybe make a new claw framework, add more functionality to market apis and potentially have it connect to my trading bot projects or other projects to make it even more useful, not only using my projects but helping me improve them. I also need to find a way to sync Fidelity portfolio to the server as currently I have to update it into my Google or Yahoo finance or have it scrape the internet from a list I give it, it works however for mutual funds like FSAGX or penny stocks like RR/BTBT it is hard to find market data directly sometimes.
Log in or sign up for Devpost to join the conversation.