TexLab is a production-grade Language Server Protocol (LSP) implementation for LaTeX and BibTeX, written in Rust. It provides the full suite of IDE features — code completion, hover documentation, go-to-definition, references, rename, document/workspace symbols, formatting, diagnostics, folding, inlay hints, document links, and highlighting — to any editor that speaks LSP (VS Code, Neovim, Emacs, etc.). Beyond editing, it orchestrates the LaTeX build toolchain: it runs latexmk, parses build logs into diagnostics, supports forward/inverse search over IPC, and integrates with chktex for linting and latexindent/tex-fmt for formatting. The architecture is a modular Rust workspace of roughly 20 crates organized around a clean separation of concerns. The `parser` crate implements hand-written recursive-descent parsers for LaTeX and BibTeX using the rowan syntax-tree library, producing typed CSTs consumed by `syntax` and `base-db`. `base-db` maintains the project workspace: open documents, a dependency graph of \input/\include/\import relations, labels, citations, environments, and .aux-derived section/equation numbering. Each LSP feature (completion, hover, definition, references, rename, symbols, folding, highlights, inlay hints, links) lives in its own crate with a uniform Params→Builder pattern, and the `texlab` binary crate wires them together via the lsp-server dispatch loop. The server communicates over stdio, with a Unix-socket IPC mechanism (with a Windows UDS fallback) for forward-search callbacks from the editor. Notable engineering investments include a deeply tested parser (thousands of snapshot tests pinning full syntax-tree shapes), a custom `citeproc` engine that renders biblatex-style formatted citations without external citation libraries, a compressed static database of thousands of LaTeX commands and packages for completion, TeX distribution detection via kpsewhich/MiKTeX FNDB binary parsing, and a declarative multi-file test fixture system reused across every feature crate. The codebase exhibits strong craftsmanship: each crate is small and focused, test coverage is extensive (snapshot tests, regression tests named after GitHub issues), and CI runs an 8-platform cross-compilation matrix with MSRV enforcement and `-D warnings`. Primary risks are inherent to the domain: regex-based build-log parsing is sensitive to TeX/latexmk output format changes, runtime behavior depends on external tools (latexmk, chktex, kpsewhich, latexindent, tex-fmt) that vary across TeX distributions, latexmkrc parsing has version-specific hacks, and large static metadata tables (BibTeX entry types, completion database) can drift from upstream CTAN/biblatex. None of these threaten the core value: this is one of the most complete and reliable LaTeX language servers available, with an architecture worth studying for anyone building language tooling.