---
title: GAIA Benchmark Agent
emoji: 🤖
colorFrom: indigo
colorTo: blue
sdk: gradio
sdk_version: 4.27.0
app_file: app_safe.py
pinned: false
hf_oauth: true
---

# GAIA Benchmark Agent

This project is an AI agent designed to tackle the GAIA benchmark, featuring multi-step reasoning, tool use (web search, Wikipedia, data analysis, file handling), and a Gradio web interface for evaluation and submission.

## Features
- LangGraph-based agent with robust tool integration
- Wikipedia, Tavily (web search), data analysis, and file handling tools
- Automatic file download for file-based questions
- Gradio interface for user interaction and answer submission
- Error handling and graceful fallback for recursion/tool loops

## Setup & Deployment

### 1. Install Dependencies
```
pip install -r requirements.txt
```

### 2. Environment Variables
Create a `.env` file (not committed) or set these variables in your Hugging Face Space:
- `OPENAI_API_KEY` (for OpenAI LLM and transcription)
- `TAVILY_API_KEY` (for Tavily web search)
- (Optional) `SPACE_ID` (for Hugging Face Space integration)

### 3. Run Locally
```
python app_safe.py
```
Or launch the Gradio interface as your main app file.

### 4. Deploy to Hugging Face Spaces
- Push your code to a public Hugging Face Space repository.
- Set your API keys as secrets in the Space settings.
- The Gradio app will launch automatically.

## Project Structure
- `app_safe.py` — Main Gradio app for full agent evaluation
- `agent.py` — Agent logic and tool orchestration
- `tools.py` — Tool definitions (Tavily, Wikipedia, data analysis, etc.)
- `requirements.txt` — All dependencies
- `README.md` — This file

## Notes
- The agent will return a fallback answer if it cannot answer within the recursion/tool call limits.
- For best results, ensure all environment variables are set and dependencies are installed.

---

**Good luck on the GAIA benchmark!**