Extract the release archive and open a terminal in the extracted directory. The engine supports standard chess and Chess960. It uses classical evaluation by default and runs offline.
Add build/caissa as a UCI engine in your chess GUI. To use it from a terminal,
start ./build/caissa and enter these commands one at a time.
uci
isready
position startpos moves e2e4 e7e5
go nodes 10000
quit
Wait for uciok, readyok and bestmove before sending the next command.
Run ./build/caissa --perft 4 for a quick installation check. The expected
result is 197281.
With Node.js 24 installed, run node dist/analysis-service/server.js from the
extracted directory. Open the local board to import a
PGN or FEN, edit variations and analyze a position. Press Ctrl+C in the
terminal to stop the service.
The bundled original network was weaker than the classical evaluator in
controlled games. To try the trained ten-percent blend in a UCI session, load
these settings after uciok.
setoption name EvalFile value models/experimental-v4.cai
setoption name Neural Weight value 10
setoption name Use Neural value true
isready
The Caissa repository has source, match records, the model card and benchmark limitations.