README.md

---
title: GAIA Unit 4 Workbench
sdk: gradio
app_file: app.py
pinned: false
hf_oauth: true
---

# Hugging Face GAIA Agent Workbench

Small Hugging Face Space/workbench for the Agents Course Unit 4 final assignment.

## What It Does

- Fetches the 20 filtered GAIA Level 1 questions from the course API.
- Downloads task files into `data/files/`.
- Lets you maintain exact-match answers in `answers.local.json`.
- Submits answers to the course scoring API from the Gradio UI or CLI.

## Required For Final Submission

- Hugging Face account.
- Public Hugging Face Space with this code.
- HF login in the Space UI.
- Exact final answers without `FINAL ANSWER` prefix.

## Local Setup

```bash
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python gaia_workbench.py fetch
python app.py
```

## CLI

```bash
python gaia_workbench.py fetch
python gaia_workbench.py download-files
python gaia_workbench.py status
python gaia_workbench.py submit --username YOUR_HF_USERNAME --agent-code https://huggingface.co/spaces/YOUR_USERNAME/YOUR_SPACE/tree/main
```

## Answer File

Create `answers.local.json` from the template:

```bash
cp answers.template.json answers.local.json
```

Then fill:

```json
{
  "task-id": "exact answer"
}
```

`answers.local.json` is gitignored because it may contain your final submission answers.