A UCI-compatible chess engine written in C++17, built from scratch with a focus on correctness and progressive optimization.
Argus is a classical chess engine that uses an alpha-beta search with iterative deepening to find the best move in a given position. It communicates via the UCI protocol, meaning it plugs directly into any major chess GUI (CuteChess, Arena, etc.) without any extra configuration.
The engine is in active development. Latest STS estimate: ~2560 Elo (v6), see Strength Estimate.
- UCI protocol — plug into any GUI out of the box, search runs on its own thread (
stopworks) - Search — negamax alpha-beta, iterative deepening with aspiration windows, principal variation search, quiescence search (SEE + delta pruning), null move pruning, late move reductions, futility and reverse futility pruning, check extension
- Time management — uses
wtime/btime/winc/binc/movestogo/movetime, orgo depth N/go infinite - Transposition table — configurable with the
Hashoption, mate-distance aware - Move ordering — TT move, MVV-LVA captures, killer moves, history heuristic, losing captures (negative SEE) sorted last
- Evaluation — tapered piece-square tables (PeSTO values), middlegame/endgame blend, pawn structure (doubled, isolated, passed pawns), piece mobility, king safety (pawn shield, open files near the king, king-zone attacks)
- Draw detection — repetition, 50-move rule, insufficient material
- Opening book — Polyglot
.bin(gm2001.bin), varied move choice - Endgame tablebases — Syzygy via Fathom (put the files in
build/syzygy/) - Legal move generation — via chess-library by Disservin
Argus/
├── lib/
│ ├── chess-library/ # Disservin's chess-library (move generation)
│ ├── fathom/ # Syzygy tablebase probing
│ └── opening-book/ # gm2001.bin Polyglot book
├── src/
│ ├── main.cpp
│ ├── engine/
│ │ ├── search.h / search.cpp # Iterative deepening, alpha-beta, qsearch, time
│ │ ├── evaluate.h / evaluate.cpp # Tapered PST + pawns, mobility, king safety
│ │ ├── movesort.h / movesort.cpp # Move ordering (TT move, MVV-LVA, killers, history)
│ │ ├── tt.h / tt.cpp # Transposition table
│ │ ├── polyglot.* # Opening book
│ │ └── syzygy.h / syzygy.cpp # Endgame tablebases (Fathom wrapper)
│ └── uci/
│ └── uci.h / uci.cpp # UCI protocol handler
├── tests/
│ ├── perft.cpp # Move generation correctness tests
│ └── sts/ # STS test suite + Elo estimate script
└── CMakeLists.txt
- GCC 15+ or Clang with C++17 support
- CMake 3.16+
- Ninja (recommended)
On Windows, use the MSYS2 UCRT64 terminal. Install dependencies with:
pacman -S mingw-w64-ucrt-x86_64-cmake mingw-w64-ucrt-x86_64-ninja
cmake -B build -G "Ninja"
cmake --build buildThe engine binary will be at build/Argus.exe (Windows) or build/Argus (Linux/macOS).
cmake --build build --target Perft
./build/PerftExpected output from the starting position:
| Depth | Nodes |
|---|---|
| 1 | 20 |
| 2 | 400 |
| 3 | 8,902 |
| 4 | 197,281 |
| 5 | 4,865,609 |
| 6 | 119,060,324 |
- Download CuteChess
- Go to Tools → Engines → Add
- Point the executable path to
build/Argus.exe - Set the working directory to
build/and the protocol to UCI - Start a game (any time control works, Argus manages its clock)
UCI options: Hash (MB), OwnBook, BookFile, SyzygyPath.
By default Argus loads gm2001.bin and the syzygy/ folder from next to the executable.
You can also talk to the engine directly in the terminal:
uci
→ id name Argus
→ id author Aitaneuh
→ uciok
isready
→ readyok
position startpos moves e2e4 e7e5
go movetime 1000
→ info depth 1 score cp 63 nodes 32 nps 0 time 0 pv g1f3
→ ...
→ info depth 13 score cp 25 nodes 6900000 nps 7500000 time 920 pv g1f3 b8c6 ...
→ bestmove g1f3
Plain go searches to depth 8, go infinite searches until stop.
tests/sts/sts.py runs the Strategic Test Suite
(1500 positions, 15 themes) and turns the score into a rough Elo estimate
(Elo ≈ 44.523 × score% − 242.85, about ±150). Build the engine first, then:
python tests/sts/sts.py # full suite (~6 min, 0.2 s per position)
python tests/sts/sts.py -n 300 # quick run, 20 positions per theme| Version | STS score | Estimated Elo |
|---|---|---|
| v3 (material only, fixed depth 6) | 31.1% | ~1140 |
| v4 (PST, qsearch, iterative deepening, time management) | 57.6% | ~2320 |
| v5 (pawn structure, king safety, mobility) | 59.9% | ~2420 |
| v6 (aspiration windows, SEE, futility pruning) | 62.9% | ~2560 |
Each new version is also run side by side with the previous one on the same machine: v4 55.7% vs v5 59.9% (+4.2), v5 59.3% vs v6 62.9% (+3.6).
- UCI communication
- Legal move generation
- Alpha-beta pruning
- Material evaluation
- MVV-LVA move ordering
- Transposition table
- Piece-square tables (PST, tapered)
- Quiescence search
- Iterative deepening
- Time management
- Null move pruning, LMR, PVS, killers, history
- Opening book (Polyglot
.bin) - Syzygy endgame tablebases (via Fathom)
- Pawn structure / king safety / mobility eval
- Aspiration windows, SEE, futility pruning
- chess-library by Disservin — header-only move generation library
- Fathom — Syzygy tablebase probing
- STS-Rating by fsmosca — STS positions and Elo formula
MIT