Start with the win-rate graph
Where the graph drops sharply is where the game turned.
Guide
No jargon needed. Download, open a game, analyze it in one click, then let the problem moves show you what mattered.
Pick the build for your computer in the official download center. On Windows, choose by graphics card: CUDA for NVIDIA RTX 20 / 30 / 40 / 50, OpenCL for AMD, Intel or older NVIDIA cards, and CPU if you have no dedicated graphics card. On a Mac, choose by chip. Unzip the Windows portable build and run it; on a Mac, open the DMG and drag the app to Applications.
Complete packages include the KataGo engine and a default network, and the first launch sets everything up. There is nothing else to download. If macOS says the app cannot be opened, go to System Settings → Privacy & Security and click Open Anyway.
Click Fox Kifu in the toolbar and enter a Fox nickname to fetch recent public games, or open an SGF file from your computer.
In the Analyze menu, choose Whole-game lightning overview to see the shape of the game quickly. For a deeper look, use Whole-game deep analysis.
Where the graph drops sharply is where the game turned.
The Problem moves list is sorted by how much each move cost. Click one to jump there, or use the arrow keys and Enter.
The colored circles on the board are KataGo’s suggestions, with win rate and visits. Play out the suggested line and compare it with the game.
Click AI Coach at the top, choose a strength, and play a practice game against AI.
Click AI Commentary at the top and sign in with ChatGPT or enter your own API key to have AI explain the position. Available models and limits depend on your provider.
Try another network, or open the GitHub backup download on the download page.
Make sure the file came from this site or GitHub, then click More info and Run anyway.
Automatic update checks are off by default. Use Help → Check update when you want one; portable users can download just the small update from the download page.
Search features in the toolbar finds menu items by name. If you are still stuck, export a diagnostics bundle from Help → Diagnostics and Logs and report the problem on GitHub Issues.