Schema Spelunker

An interactive desktop tool for exploring database schemas — fast, native, GPU-accelerated, and built in the handmade spirit.

Inspired by SchemaSpy and its wonderful diagramming, but aiming for a live, interactive experience rather than static HTML output.

Vision

• Open a database file in a GUI and immediately see its schema
• ER diagrams rendered interactively — zoom, pan, select, explore
• Sub-diagrams showing 1–2 degrees of separation from a selected table
• Everything stays fast because it runs natively on the GPU
• Dump the schema to a custom snapshot format so you can explore offline without repeatedly hitting the database
• CLI mode (schema_spelunker path/to/database.db) for scripting and CI

Philosophy

Handmade software in the spirit of Casey Muratori, Ryan Fleury, and gingerBill: minimal dependencies, own the stack where it counts, no framework magic. The project uses Odin for its no-nonsense approach to systems programming and its excellent FFI for binding C libraries like SQLite.

Optimisations are driven by profiles, not hunches. The codebase builds things simply first and only adds complexity when real-world usage proves it matters. String interning, #soa layouts, and other cache-oriented tricks live in the "If needed later" section of TODO.md until a profile says otherwise.

Current status

Early but building fast.

Stack: SDL3 (window + input), OpenGL 3.3 core (rendering), Dear ImGui (docking branch, C-ABI via dcimgui), ImNodes (ER diagram), SQLite (amalgamation, statically linked). Borderless window with an owner-drawn titlebar and optional OS glass blur on Windows.

SQLite: vendored as source (vendor/sqlite3/sqlite3.c amalgamation) instead of a binary .dll, compiled to a static .lib on the first build. Schema introspection via PRAGMA table_info / PRAGMA foreign_key_list.

Build: all native dependencies (SQLite, ImGui, ImNodes, SDL3 backends) are compiled from source on the first build via MSVC (auto-detected with vswhere) or clang/mold in the Nix dev shell. No binary blobs in the repo.

Project layout

main.odin            app entry point, GUI loop, CLI schema dump
style.odin           theme data + .ssTheme loader, ImGui/ImNodes styling
blur_windows.odin    Windows glass blur (#+build windows)
blur.odin            no-op blur shim for other platforms
assets/              runtime resources: Roboto.ttf, icons/, themes/
vendor/              vendored C sources + Odin bindings (sqlite3, imgui,
                     sdl3_headers, gl, stb)
docs/                design docs + C_STYLE_GUIDE.md, ODIN_STYLE_GUIDE.md
test/                seed tools and SQL schema
tools/               asset generators (gen_icons.odin)
build.bat, build.sh  Windows / Linux build scripts

Building & running

Requires MSVC (Visual Studio 2022 Build Tools or newer) on Windows. The build script auto-detects your VS installation via vswhere.

build.bat              # debug build (compiles native libs on first run)
build.bat run          # debug build + run
build.bat release      # optimized build (-o:speed)
build.bat debug        # debug build with symbols (-o:none -debug)
build.bat clean        # remove build output

Flags can be combined: build.bat debug run, build.bat release run.

Linux uses the same interface via build.sh:

./build.sh run          # debug build + run
./build.sh release      # optimized build (-o:speed)
./build.sh debug        # debug build with symbols (-o:none -debug)
./build.sh clean        # remove build output

build.sh debug run and build.sh run debug both work.

The spelunker takes a database path as a CLI argument:

schema_spelunker.exe path/to/database.db

Running without arguments opens the GUI.

Seeding a test database

A sample multi-tenant Kanban board schema is in test/complex.sql with 16 tables, users, teams, roles, permissions, boards, columns, cards, labels, comments, and activity history.

seed.bat                         # creates seed.db from test/complex.sql
schema_spelunker.exe seed.db     # explore it via CLI
build.bat run                    # explore it via GUI