I Have 1 Hour Free" A Walk Plan in 10 Seconds, Powered by Local AI

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

Free Time Walk Planner is a local AI agent that turns "I have 1 hour free" into a real plan for getting outside.

It picks a nearby park, tells you when to leave and when to turn back so you're home before sunset, and says what to carry based on the weather.

It's for anyone with a short gap in their day who would otherwise spend it scrolling. The answer takes about 10 seconds to read, and then you're out the door. The screen is the shortest part of the experience.

Tell the agent how much free time you have and where you are.

A short plan you can read in about 10 seconds: where to go, when to leave, and when to turn back.

Nearby parks and trails, pulled from OpenStreetMap.

Full transparency: every tool the local models chose, and what each one returned.

Code

abhisek247767 / free-time-agent

A local-first AI agent that turns your free time into a concrete outdoor plan using Ollama, LangChain, weather, sunset, and nearby green-space tools.

View on GitHub

How I Built It

Two small local models on Ollama, each with one job:

• tev1:0.8b: a tiny decision model that only decides which tools a request needs (plan a walk, weather, daylight, or politely decline).
• qwen2.5:3b: fills in the tool arguments via LangChain tool calling, then writes a short, friendly headline.

Free, keyless open data:

• Open-Meteo: weather forecast
• Nominatim (OpenStreetMap): place name → coordinates
• Overpass API (OpenStreetMap): nearby parks, gardens and trails
• OSRM: walking routes and distances
• astral: sunrise, sunset and golden hour, calculated fully offline

Safety logic in plain Python, not the LLM:
Leave-by, turn-back and home-by times, plus warnings for rain, wind or a return close to sunset, are all calculated in code, so the numbers are always correct. The LLM only explains them. Tool results are cached locally, so a plan still works when the network drops.

Frontend:
A Streamlit app with live progress, a map of nearby parks, and a "how this answer was made" view that shows every tool call the agent made.

Why Does Open Innovation Matter?

• Private by design: Both models run on my machine. My location and walking habits are never sent to a closed AI provider.
• Zero cost: No API keys, no usage bills. A daily "go outside" nudge costs nothing.
• Right-sized models: Open weights let me pair a 0.8B model for decisions with a 3B model for writing. That split is fast on a laptop, and I can swap either model without touching the rest.
• Open data for the real world: OpenStreetMap and Open-Meteo give the agent real parks and real weather for free.

A closed API would have turned a personal "go touch grass" tool into something that costs money per request and sends my location to someone else's server.

Prize Categories

• Best Use of Gemma (Featured categories)
• Best Use of GitHub Copilot (Partner categories)