DataSkein

Follow the thread in your data without sending the source files away.

Open DataSkein · Read the product decision · Read the naming decision · See the roadmap

DataSkein is a focused, local-first workbench for recurring CSV, JSON, JSONL, Parquet, and DuckDB data. Open flat files or inspect a local DuckDB table in the browser, shape the result, create a compact chart, and export an auditable recipe.

It is deliberately not a spreadsheet, a general BI suite, a cloud data platform, or an AI analyst.

Why this exists

People repeatedly reach for a spreadsheet to inspect an export, then hit row limits, browser stalls, manual joins, or a transformation process that is hard to explain the next day. Browser-local DuckDB tools prove that local analysis is useful, but the generic SQL workbench category is already crowded.

DataSkein's narrower job is:

Turn a recurring set of awkward local exports into a small, visible recipe and a result another person can inspect.

The evidence, competitor comparison, and scope decision are documented in docs/PRODUCT.md.

Release scope

• Drag and drop CSV, TSV, JSON, JSONL, NDJSON, Parquet, and DuckDB files
• Read-only DuckDB catalogs with schemas, tables, views, column counts, and bounded base-table inspection
• Content checks that reject obvious extension spoofing before parsing
• A local CSV import X-ray with explicit delimiter, header, encoding, and all-text recovery
• Local schema and column profiling for flat files with bounded previews
• Visual filters, sorting, column selection, and left or inner joins
• Inspectable DuckDB SQL generated from every transformation step
• Bar, line, and area charts with accessible descriptions
• Small in-session dashboards
• Formula-neutralized CSV export
• Portable SQL recipe export with source names and SHA-256 fingerprints
• Standalone HTML dashboard export containing only aggregated chart data
• Stage-based local import and query progress with contextual recovery actions
• A resumable first-result guide and explicit loaded-session offline state
• A single browser worker, a 1 GB per-file limit, and a 1 GB query memory cap
• No account, workspace analytics, cookies, upload endpoint, or remote database connector

Source files are registered with DuckDB-Wasm through the browser File API. DataSkein does not intentionally transmit them. See SECURITY.md for the exact trust boundary and docs/ARCHITECTURE.md for the data flow.

The separate public landing page uses our self-hosted Umami instance for aggregate page, bounded CTA, section, scroll-depth, and engaged-time events. It respects Do Not Track and Global Privacy Control, sets no analytics cookies, and never receives source-file contents or workspace actions. URLs are reduced to the landing root plus safe, bounded standard UTM values, and referrer paths are removed before sending.

Quick start

Requirements: Node.js 22 or newer and pnpm 11.7.

pnpm install --frozen-lockfile
pnpm dev

Then open http://localhost:5173. Use the built-in sample or drop your own files into the workspace.

Quality gates

pnpm check
pnpm test:e2e
pnpm audit
pnpm release:artifact

The E2E suite covers the complete sample journey, JSON and JSONL, an Apache Parquet interoperability fixture, a checkpointed DuckDB database with 100,000 rows, semicolon, headerless, and Latin-1 CSVs, a 64 MiB CSV, malformed input, file-type spoofing, spreadsheet-formula neutralization, accessibility, responsive layout, offline behavior, mobile console and overflow checks, and the absence of cross-origin requests during local source workflows.

More detail is in docs/DEVELOPMENT.md.

Browser and size limits

DataSkein targets current Chromium, Firefox, and Safari-capable browsers, while CI currently exercises Chromium. The app refuses individual files above 1 GB. This is a safety boundary, not a claim that every 1 GB workload will succeed. DuckDB-Wasm shares the browser's WebAssembly memory and storage constraints, so wide schemas, deeply nested JSON, expensive joins, or several large files may fail earlier. The UI caps previews at 250 rows and keeps query work off the main thread. DuckDB tables do not receive an automatic full row count or column profile. Persisted views are listed but not executed because their stored SQL may reference external sources.

Contributing

Issues and small, test-backed pull requests are welcome. Start with CONTRIBUTING.md, the roadmap, and the code of conduct.

License

MIT © 2026 Artur Ferreira.