README.md

---
title: Agent 101
emoji: 🤖
colorFrom: indigo
colorTo: pink
sdk: gradio
sdk_version: 5.9.1
app_file: app.py
python_version: "3.11"
pinned: false
license: apache-2.0
short_description: LLM side-by-side with and without tool access.
---

# Agent 101

A tiny demo to show what "giving an LLM tools" actually does.

Both sides of the app call the **same model** via the HF Inference Providers
API (default: `meta-llama/Llama-3.1-8B-Instruct`). The left side runs the
model plain. The right side hands it a small toolkit (calculator, weather
lookup, unit converter, course-notes search, CS/ML dictionary) and runs an
agent loop.

Try a question the plain LLM can't answer — e.g. `What is 4729 times 8314?`
or `How much hotter is Delhi than Bangalore right now?`. The plain model
will guess or refuse; the agent will reach for the right tool, execute it,
and answer from the real data.

Because inference is delegated to HF Inference Providers, this Space runs
fine on **free CPU Basic** hardware — no GPU required.

Based on the Week 12 lab for CS 203 (Software Tools and Techniques for AI)
at IIT Gandhinagar. The full Colab is at
[`lecture-demos/week12/colab-notebooks/01-agents-from-scratch.ipynb`](https://github.com/nipunbatra/stt-ai-teaching/blob/master/lecture-demos/week12/colab-notebooks/01-agents-from-scratch.ipynb)
in the course repo.

## Running locally

```bash
pip install -r requirements.txt
export HF_TOKEN=hf_...
python app.py
```

To swap the backing model:

```bash
export MODEL_ID="Qwen/Qwen2.5-72B-Instruct"   # or any tool-capable chat model
```

## Deploy to HF Spaces

```bash
git init
git remote add origin https://huggingface.co/spaces/Nipun/agent-101
git add .
git commit -m "Initial commit: Agent 101"
git push -u origin main
```

Then set the `HF_TOKEN` secret in the Space's Settings → Variables and Secrets.
CPU Basic (free) is enough — no GPU needed.