Getting Started with Xiangqi AI and Game Analysis
Where Should You Start on a Phone or Computer?
| Your situation | Start here |
|---|---|
| You want to try it without learning how to install and load an engine | See the web version guide, for phones and computers. |
| You want to install Xiangqi software for your system | See Download Pikafish and choose Windows, Android, iOS, macOS, Linux, or HarmonyOS. |
| You already use the Pikafish app | See App Help for each device's import, analysis, and save controls. |
| You already have a familiar GUI and know how to load engines | Download the engine from the official website, then configure it in your GUI. |
Get analysis working on one position before adjusting advanced settings. The web version downloads the engine on first open; an unfinished download doesn't mean it can't analyze.
What Is the Difference Between an Engine and a GUI?
The engine calculates moves; the interface is how you use it. Pikafish is a Xiangqi engine; the board, move list, import buttons, and analysis panel usually come from the graphical user interface (GUI).
Apps that combine both work as soon as you open them; an engine-only download must be loaded into a compatible GUI, so seeing no board after downloading an engine doesn't necessarily mean anything is broken. See What Is an Engine? and What Is a GUI?.
Review a Game with the Engine
These steps use the web version; for the app, see its help page.
- Import the game. Choose a game file, paste the moves, or play the game out yourself. Image recognition gives only the current position, not the earlier moves; check the pieces and side to move afterward. See Import a Game.
- Scan the whole game first. Select “Game analysis” at the bottom of the move list (on a phone the button reads “Review”), then choose “Fast,” “Standard” or “Thorough.” “Thorough” takes the longest and also marks brilliant and great moves. Phone and computer controls: see Game Analysis.
- Return to the position before a problem move. Compare the recommended move with your actual move, and follow the suggested line a few plies to see where they differ. Understand the missed defense, lost material, or missed attack first; don't rush to memorize a long line.
- Write your own explanation. Add variations and notes (for example, “Defending first stops the opponent from capturing with tempo”), then save or export the game. See Variations and Notes.
Playing a move or leaving the page stops whole-game analysis; completed results are kept. To try moves yourself, wait until it finishes, then go back through the game.
How Do You Read Scores, Depth, and PV?
Check whose perspective the score uses. The web version labels it “Red” or “Black”; a positive value favors the named side. Other GUIs may differ. A score is the engine's assessment, not a guaranteed result, and scores from different versions shouldn't be compared directly.
Depth does not mean every line was searched equally far. The engine prunes, reduces, or extends different branches. Scores and recommended moves may change early in analysis; give key positions more time rather than watching how fast the depth climbs. See Depth.
PV is the currently recommended principal variation. It includes both sides' moves and is the engine's current main line; the opponent won't necessarily follow it. If they choose a different reply, analyze that position again.
For where scores come from, see Position Evaluation and NNUE; for UCI cp, mate, and WDL, see Score Output.
What Should You Check When Results Differ or Analysis Does Not Respond?
- Is the position correct? Check the pieces, side to move, and that editing is finished. For perpetual check or chase, also keep the starting position and move history; a FEN alone can't restore the whole history.
- Which result are you looking at? In the web version, “Engine” calculates on your device, while “ChessDB” looks up stored data, which may not include the current position; see ChessDB.
- Are the settings appropriate? Start with defaults. Before adjusting threads, Hash, or the number of variations, check UCI Options; larger values aren't automatically better.
Local analysis uses power and generates heat; turn it off when done. In each review, understand a few key lines, then think them through on the board yourself to turn AI suggestions into your own judgment.
