logo

Steam Buzz

HomeAboutSupport me

Steam Buzz is a hobby project and is not affiliated with Valve or Steam.
Steam and the Steam logo are trademarks of Valve Corporation. All other trademarks are property of their respective owners.

Project Summary

Steam Reviews AI

is a Nextjs web application that analyzes and summarizes user reviews for Steam games using a cloud LLM. It fetches reviews directly from the Steam API, checks previous history in a cloud database, and generates concise summaries to help users quickly understand the strengths and weaknesses of any game.

How it Works

  • Game Search:

    - Gets the current list of all games from the Steam API: Steam API
    - Searches by title using fuzzy matching for similarity.
    - Shows up to 30 most likely games, with review stats and icons.

  • Game Analysis:

    - When you select a game, a banner and its stats appear.
    - Clicking "analysis" fetches the top 20 English reviews: Steam Review Docs
    - These reviews are sent to Mistral AI for summarization.
    - If a recent summary exists in the Neon database, it is reused for speed.

  • Summary Output:

    - The agent summarizes the reviews, infers a description, gives a score (0–10), and lists positive/negative factors.
    - Results are shown as images and markdown for easy reading.
    - Users can report issues to help improve future summaries.

Why?

I've always wanted a quick way to see what a game is about and what Steam users really think—especially the negative opinions. I would love for Steam to have something like this. Often I try reading reviews to find out why a game had a low or high score, but it takes too long to find out. This project also helped me practice Streamlit, set up my first cloud database, and experiment with agentic AI. One of the goals of this project is to try getting Valve attention to add a feature like this to Steam.

I have a previous project that tries doing something similar with no LLM involved. You can check it here: Streamlit App or GitHub Repo. The results are fun but it feels a bit like going to an oracle that speaks in riddles... I also implemented pretty much this same thing with Streamlit (Python) first. You can check that on my github repos.

Setup

  • Hosting: Vercel (free tier)
  • Database: Neon (cloud, free tier)
  • AI Provider: Mistral (model: small-latest, pay as you go, could have gone with free + queues)
  • Tech: Nextjs and Chakra-ui plus some standard js libraries

You may encounter rate limits due to free tiers.

About the Developer

This project was created by Vicente Arce David, alias "Duerkos".

GitHub: github.com/Duerkos
Landing Page: duerkos.github.io
Ko-fi: ko-fi.com/duerkos