Kibitz

A chess coach and database that explains positions in human terms.

Most chess software will tell you that a move is −1.3. Kibitz tells you why: which imbalances define the position, whose plan is faster, which piece is loose, and what you keep getting wrong across your own games — then trains you on exactly those weaknesses.

Kibitz is a free, open-source desktop app (macOS and Linux). Project site: kibitzchess.org — including the rendered user guide. It is built around three ideas:

1. Explain, don't just evaluate. A static explanation engine reads the position the way a teacher would — material, pawn structure, king safety, piece activity, files and diagonals, space, development — and renders the verdict as prose with evidence you can click. The teaching style is inspired by Jeremy Silman's imbalance framework.
2. Train on your games. Kibitz profiles your imported games (motif matrix, structure report, phase accuracy, conversion/defence) and feeds the findings straight into its trainers. Every claim links to the exact ply in the exact game that produced it.
3. The engine stays off by default. Stockfish runs only when a tactical screen says the position is sharp, when you explicitly ask, or in batch jobs you start yourself. Quiet positions get quiet, human answers.

Taken from a pre-release build; the window still shows the project's working title in place of "Kibitz".

Profile — every number is a claim, every claim opens the game that produced it:

Endgame trainer — moves graded against Syzygy tablebases, never an engine score:

Tactics — weakness-targeted, rated, Woodpecker cycles, speed drills:

Coach

• Static explanations of any position: imbalances, plans, tactical alerts, in a coach or neutral voice — offline, no engine, no network.
• Delta narration: what the last move actually changed.
• Player profile built from your own games: motif matrix (missed vs. allowed), structure scores, phase accuracy (ACPL), conversion and defence rates — with click-through evidence to the source ply.
• Opponent prep: fingerprint an opponent's games, find the weak lines, and turn them into a prep sheet.

Train

• Repertoire trainer: per-color opening repertoires scheduled with FSRS-4.5 spaced repetition.
• Tactics trainer: rated puzzles from the Lichess puzzle database (CC0), weakness-weighted from your profile, plus motif filters, Woodpecker cycles, and speed drills.
• Endgame trainer: a rating-tiered curriculum where every move is graded against Syzygy tablebases — still winning, slower (with the DTZ cost), or throws the win — never a bare engine number.

Database

• SQLite-backed personal database with import from SCID (.si4) and PGN; PGN export for round-tripping.
• TWIC ingest (downloads to your machine — TWIC data is never bundled).
• Lichess and chess.com account sync, FICS archives.
• Opening tree with ECO names, position search, duplicate detection with source-aware precedence.
• Batch annotate/analyze through a job queue — resumable, pausable, with honest time estimates.

Download the latest release for your platform from GitHub Releases.

• macOS: download the .dmg, drag Kibitz to Applications.
• Linux: download the .AppImage (make it executable) or the .deb.

To use engine analysis, point Kibitz at a Stockfish binary in Settings → Engine (or leave the path empty to let Kibitz resolve one automatically). For tablebase-verified endgame training, point Settings at a local Syzygy tablebase directory. Both are optional — everything else works offline out of the box.

Prerequisites:

• Rust stable, 1.82 or newer (rustup)
• Node.js 22 or newer, with npm
• Linux only: Tauri's system libraries — libwebkit2gtk-4.1-dev build-essential curl wget file libxdo-dev libssl-dev libayatana-appindicator3-dev librsvg2-dev

git clone https://github.com/avienu/kibitz.git
cd kibitz

# Core crates + database layer
cargo test --workspace

# Front end
cd app
npm install
npm test

# Run the desktop app in dev mode
npm run tauri dev

# Or produce a release bundle (app/src-tauri/target/release/bundle/)
npm run tauri build

• SCID: open Import PGN / SCID and point it at a .si4 database; games, names, and headers import directly.
• PGN: import any .pgn file, or paste PGN straight into the import view.
• ChessBase: native ChessBase formats are out of scope — export your database to PGN in ChessBase, then import the PGN.
• Online play: connect your Lichess and chess.com accounts under Account syncs; ingest TWIC issues under TWIC ingest.

Every imported dataset's provenance (source, license, date) is tracked in the database.

Kibitz is split by design into two license layers:

• crates/* (explanation engine, profiling, SRS scheduling, tablebase probing, SCID reading) — BSD-3-Clause. These crates are free of GPL dependencies, enforced by a CI license gate.
• app/ (the Tauri desktop app and its database layer) — GPL-3.0.

See LICENSE-BSD, LICENSE-GPL, the per-crate LICENSE files, the dependency registry in docs/LICENSES.md, and THIRD-PARTY-NOTICES.md.

• Jeremy Silman (1954–2023), whose imbalance-based teaching in How to Reassess Your Chess inspired the explanation engine's approach. Kibitz is not affiliated with or endorsed by his estate.
• Stockfish (GPL-3.0), run as a separate user-provided process for tactical screening and analysis.
• Lichess for the CC0 puzzle database and openings dataset.
• Fathom (MIT) for Syzygy tablebase probing, vendored in crates/kibitz-tb.
• chessground (GPL-3.0) for the board UI, and Colin M.L. Burnett's classic cburnett piece set bundled with it.
• cozy-chess (MIT) for board representation and move generation.