A local-first, schema-driven knowledge base for manufacturing engineers. PDFs go in; highlighted, source-backed entries come out. Fully offline, single self-contained binary per OS, MIT-licensed.
Specs and codes — BAC3082, AMS-C-5541, internal part numbers — live
buried in PDFs across SharePoint and network drives. Engineers spend ten
minutes per code hunting for definitions; the codes drift across spellings;
tribal knowledge walks out the door with retirees.
Specdex lets a team curate one or more knowledge bases of these codes — each with its own schema — by highlighting text in source PDFs. After curation, every lookup is a sub-second database hit, and new documents are auto-highlighted when they reference known codes.
v1 is single-user, single-machine, fully offline, no AI. v1.1 adds LAN team-mode via shared-secret auth (~3 weekends of additive work on top of v1's API-shaped backend).
- No AI. No telemetry. No outbound network requests.
- No fuzzy matching — aliases are explicit.
- MIT license; no GPL/AGPL/LGPL dependencies linked into the binary.
- Single self-contained binary per OS — no Python/Java/Node runtime needed.
- Your data is portable: full backup as ZIP, per-KB JSON export.
Prebuilt installers aren't published yet — for now, build from source
(see Build from source below). Signed, notarized per-OS
installers (.msi / .dmg / .AppImage) are planned once code-signing is set up.
- Open Specdex.
- The onboarding wizard asks for your display name and walks you through creating your first knowledge base from a template (Boeing Specs, Material Codes, Internal Part Numbers, or Empty).
- Drag PDFs onto the app. Specdex parses them and scans for codes you've already defined.
Three starter KBs ship under examples/kbs/. Import via Settings → "Import
KB JSON" or KB list → "Import" once Specdex is open.
git clone https://github.com/bsherifi/specdex
cd specdex
pnpm install
./installers/download-pdfium.sh linux # or macos-arm, windows-x64
./installers/download-ocrs-models.sh
pnpm tauri:dev # dev mode
pnpm tauri build # production bundle for this OSRequires Rust 1.85+ (managed via rustup; pinned in rust-toolchain.toml)
and Node 22+.
See CONTRIBUTING.md. The architecture is API-shaped —
business logic lives in specdex_core behind a thin Tauri adapter — so most
features land as one core function + one Tauri command + one React route.
Test coverage is non-negotiable; CI gates on cargo test, pnpm test,
cargo clippy -- -D warnings, pnpm lint, and Playwright E2E on Linux.
| Version | Highlights |
|---|---|
| v1 (current) | Single user. PDF only. Solo workflow. |
| v1.1 | LAN team-mode (host/client/solo). DOCX + XLSX. Region capture. OCR text-overlay. |
| v2 | Global hotkey. Browser extension. Folder watch. Opt-in LLM assist for non-primary fields. |
MIT. Third-party licenses bundled with the binary at
THIRD-PARTY-LICENSES.txt.
- pdfium-render + PDFium (Apache 2.0 / BSD-3) for PDF parsing.
- ocrs (MIT / Apache 2.0) for pure-Rust OCR.
- Tantivy (MIT) for full-text search.
- Tauri (MIT/Apache 2.0) for the desktop shell.
- shadcn/ui (MIT) and Radix UI (MIT) for the UI primitives.