Reads every Steam review of a game on your own computer: what players praise and criticise, how often, and the reviews behind every figure. A desktop app, with a command line and an MCP server for AI assistants.
星數
2
工具
—
可用率
—
A tool for finding out what players of a game actually think, rather than what the loudest
reviews say.
Install
The installer and the portable archive for each system are in the
latest release. A package manager
installs it too:
# Windows, with winget: the installer
winget install Aureliolo.SteamGauge
# Windows, with Scoop: the portable program, with a Start menu shortcut
scoop install https://github.com/Aureliolo/steamgauge/releases/latest/download/steamgauge.json
# macOS on Apple Silicon, with Homebrew: this repository is the tap
brew tap aureliolo/steamgauge https://github.com/Aureliolo/steamgauge
brew install --cask aureliolo/steamgauge/steamgauge
Install has the rest: the installer and archive for each system, what each
system needs, the first launch, verifying a download and upgrading.
Install: the files for each system, the first launch, checking a download and
updating.
How it works: what it downloads, how it reads a review, and what each
figure counts.
Accuracy: how often the reader is right, against what, in which languages, and
what the figures cannot tell you.
AI assistants: the app's MCP server and the assistants it adds itself to.
MCP tools: every prompt and tool the MCP server offers and what each is given.
Command line: every command of the steamgauge program.
Architecture: how the parts fit and the rules they keep, for anyone
changing SteamGauge.
Privacy: what stays on your computer and the four places the app connects to.
Models and dataset
The readers, the search models and the labelled claims are on Hugging Face, each release
tagged v1, v2 and so on. The app fetches every file from one pinned commit and checks its
length and SHA-256 before using it.
The search models are fp16 ONNX exports of Qwen3-Embedding-0.6B and Qwen3-Reranker-0.6B with
their weights unchanged. The writer is Qwen3-4B-Instruct-2507 with its weights rounded to four
bits. The dataset holds no review text and nothing about the authors: each
labelled review carries a SHA-256 fingerprint of its text, and training/fetch_text.py
fetches the reviews from Steam and puts the text back on your own machine.
Reads every Steam review of a game on your own computer: what players praise and criticise, how often, and the reviews behind every figure. A desktop app, with a command line and an MCP server for AI
Reads every Steam review of a game on your own computer: what players praise and criticise, how often, and the reviews behind every figure. A desktop app, with a command line and an MCP server for AI assistants.
上次檢查時間:2026-11-15。 Reads every Steam review of a game on your own computer: what players praise and criticise, how often, and the reviews behind every figure. A desktop app, with a command line and an MCP server for AI assistants.