Touch Grass

Inspiration

In today's fast paced world it can be easy to get caught up in a tidal wave of assigments, deadlines, and obligations. Touch Grass was designed as a way to make it easy to get out of the house by taking all the hassle out of planning a day, night, or afternoon on the town

What it does

Touch Grass leverages the latest in agentic AI technology to dynamically plan the perfect day, date, or weekend by scanning events, restaurants, landmarks, and activities in your immediate vicinity.

Use AI to create the perfect day from scratch, or let our agents make sure you aren't missing any critical parts of your already-planned day. Did you remember to find a restaurant for dinner? What about checking if that museum is closed on Mondays?

How we built it

Front End - Next.js build with Lovable, sculpted by our team of experts Backend API - FastAPI (Python) Database / Auth - Supabase Agents - LangChain / Agentnuity Cloud / Hosting - Render

Challenges we ran into

Dealing with python dependencies was really tough, especially considering LangChain's confusing documentation.

Agenuity worked great, but one we deployed it we wouldn't figure out how to get responses from the agent back to our front end (something to do with web hooks)

GPT 5 was incredibly slow, but GPT 4o while faster, doesn't generate as good output so we had to find a happy balance.

LangChain's internet search tool doesn't work like it appears in the docs

Accomplishments that we're proud of

Getting something hosted and talking to each other!

What we learned

Difference between LangChain and LangGraph

What a diffusion model is

What's next for Touch Grass

We missed out on tons of features that could have really made this application cooler. We feel all the plumbing is there, and we'd just need to clean up our agent implementation and add a couple more tools so the agent can pull event data and leverage geo location.

Built With

• agentnuity
• fastapi
• langchain
• next.js
• render
• supabase