Team Ai
Apppublic

csongorosz/gaia-langgraph-agent

sourceHugging Faceapache-2.0updated 2mo agoView on Hugging Face
0likes
App README

GAIA LangGraph Agent

A local-first AI agent built with LangGraph for the Hugging Face Agents Course GAIA evaluation.

Features

  • —ReAct-style agent loop implemented with LangGraph
  • —Gemini and local Ollama model support
  • —Web search and webpage extraction
  • —Calculator and text-processing tools
  • —Python/text attachment inspection
  • —Excel spreadsheet analysis
  • —Chess-position analysis
  • —Image and audio attachment support
  • —Local answer persistence for GAIA submissions
  • —Automated tests for tools and evaluation logic

Setup

Requires Python 3.12 and uv.

powershell
uv sync --dev
Copy-Item .env.example .env

Add your Gemini API key to the private .env file. The .env file is excluded from Git.

For local Ollama usage:

powershell
ollama pull qwen3:8b

Usage

powershell
uv run gaia-langgraph-agent demo
uv run gaia-langgraph-agent ask "What city is Hugging Face headquartered in?"
uv run gaia-langgraph-agent questions
uv run gaia-langgraph-agent download-files
uv run gaia-langgraph-agent gaia <task-id>

Tests

powershell
uv run pytest

Project goal

This project was created for Unit 4 of the Hugging Face Agents Course. It demonstrates a testable, tool-using LangGraph agent capable of handling multimodal GAIA benchmark tasks.