Tableau stores your business logic — field definitions, calculations, dashboard layouts, filter wiring — in formats that no AI system can read. drexo changes that.

Point drexo analyze at any .twbx file and get back two structured markdown files: a complete metadata model and a visual blueprint with SQL query specifications. Pass them to any LLM, feed them into a code generator, or use them to scaffold a React frontend backed by a real API.

How It Works

Tableau Workbook (.twbx)
        │
        ▼
   drexo analyze [--enrich]
        │
        ├──▶  .model.md    — metadata: fields, calculations, data sources,
        │                    worksheets, dashboards, filters (markdown + YAML)
        │
        └──▶  .visual.md   — visual blueprint: dashboard layout, per-worksheet
                             SQL query specs, interactive filter wiring,
                             field dictionary, calculated field translations

With --enrich, both files pass through Claude for a second pass: AI-generated executive summary, worksheet descriptions, plain-English formula translations, and cleaned field names with business context added to every SQL block.

What's in the Output

.model.md — Structural Metadata

Everything drexo extracts from the Tableau XML:

• Workbook metadata — name, Tableau version, export timestamp
• Data sources — connection type, all fields with types and roles
• Semantic layer — every field: raw name, display caption, data type, role
• Calculated fields — Tableau formulas captured verbatim, with AI-simplified versions when --enrich is used
• Worksheets — mark types, AI-generated descriptions
• Visual encodings — rows shelf, columns shelf, color, size, filters per worksheet
• Dashboard layout — full zone tree with coordinates preserved for grid reconstruction
• Filters & actions — all cross-sheet action filter wiring
• Executive summary — AI-generated when --enrich is used

.visual.md — Visual Blueprint

Designed for developers and AI agents building the frontend and backend:

• Dashboard layout — proportional ASCII layout + table (61%/39% column splits, etc.)
• Per-worksheet query specs — SQL equivalent of what Tableau executes, with filter parameters
• Interactive wiring — which fields connect which worksheets as action filters, and the URL param name for each
• Calculated field translations — Tableau IF/THEN/ELSE → SQL CASE WHEN
• Field dictionary — internal Tableau names → clean display names
• Data source summary — base columns with types and roles

Quick Start

npm install -g drexo

Analyze a single workbook

drexo analyze ./my-report.twbx
# Writes: my-report.model.md + my-report.visual.md

AI-enriched output

export ANTHROPIC_API_KEY=sk-ant-...
drexo analyze ./my-report.twbx --enrich
# Claude enriches both files: summaries, descriptions, cleaned SQL
# Results are cached — repeat runs are free

Analyze an entire folder

drexo analyze ./workbooks/ --output-dir ./reports
# Finds all .twbx files, runs 3 in parallel
# Writes organized subdirectories + index.md catalog

Explicit batch

drexo analyze report-a.twbx report-b.twbx --output-dir ./reports --enrich

Commands

| Command | Description | |---|---| | drexo analyze <files...> | Analyze one or more workbooks. Accepts files, a directory, or a mix. | | --output-dir <path> | Write outputs into organized subdirectories. Generates an index.md catalog when multiple workbooks are processed. | | --enrich | Enrich output with AI-generated summaries, descriptions, and cleaned SQL (requires ANTHROPIC_API_KEY). Results cached by content hash. | | --enrich-model <model> | Claude model to use (default: claude-haiku-4-5-20251001) | | drexo --debug | Enable debug logging (stack traces on errors) | | drexo --help | Show all options and examples | | drexo --version | Print the drexo version |

Output structure with --output-dir

reports/
  index.md                              ← catalog: all workbooks, worksheet counts, summaries
  accounts-receivable-analysis/
    model.md
    visual.md
  superstore-analysis/
    model.md
    visual.md

Why Two Files?

.model.md answers: "what is in this workbook?" .visual.md answers: "how do I rebuild it?"

The visual blueprint is the document a developer or AI agent needs to generate both the React frontend components and the backend API routes. It maps directly to code:

• Dashboard layout → CSS grid percentages
• Query spec → API endpoint + SQL
• Filter wiring → shared URL query params or React context
• Field dictionary → TypeScript interface names

Roadmap

| Phase | Status | Goal | |-------|--------|------| | v0.1 — Metadata Wedge | ✅ Shipped | drexo analyze → .model.md | | v0.2 — Visual Blueprint + AI Enrichment | ✅ Shipped | --enrich, .visual.md with SQL specs, batch processing, organized output | | v0.3 — React Generator | Next | drexo migrate → Vite + React app wired to real API endpoints | | v1.0 — Production-Ready | Planned | Live data, interactive filters, calculated fields in generated code | | post-v1 — Data Agents | Conditional | drexo query — conversational Q&A over your BI catalog |

Full roadmap: docs/ROADMAP.md

Development

git clone https://github.com/raguvindtharanitharan/drexo.git
cd drexo
npm install

# Run the CLI without a build step
npm run dev -- analyze ./examples/Accounts\ Receivable\ Analysis.twbx

# Build
npm run build

# Run tests
npm test

Tech stack:

• ESM-only TypeScript — no CJS shims, no __dirname hacks
• commander for the CLI surface
• fast-xml-parser for .twb parsing; adm-zip for .twbx unzip
• @anthropic-ai/sdk for the enrichment layer
• Vitest + real-fixture integration tests (no mocked XML)

Contributing

The parser and visual blueprint generator are the highest-leverage areas right now. Every new workbook surfaces edge cases that make drexo more generic.

• Read CONTRIBUTING.md
• Issues labeled good first issue and parser are great starting points
• Real workbooks (even anonymized) are the most valuable contribution

License

MIT © Raguvind Tharanitharan

Made for teams tired of vendor lock-in. If drexo saves your company money, star the repo and tell your friends.