A TypeScript port of pandas, built from first principles using Autoloop — an automated research and experimentation platform that runs iterative optimization loops on GitHub Agentic Workflows.
🎮 Try the interactive playground →
Pronounced tseh-SEH-bee · IPA /tsɛˈsɛbi/ — named after the tsessebe (Damaliscus lunatus), a southern African antelope. The word is Setswana; it carries the "ts" as one sound, the middle "se" takes the stress, and the final "be" is unstressed.
- Package name:
tsb— all imports and usage usetsb, nottsessebe.import { DataFrame } from 'tsb' - Runtime & tooling: Bun for everything — runtime, bundler, test runner, package manager
- Language: TypeScript in strictest mode — no
any, noascasts, no@ts-ignore, no escape hatches - Dependencies: Zero for core library. External deps only where absolutely required for non-core tooling (e.g. Playwright, WASM toolchains).
- Linting: Biome with all rules enabled, zero warnings tolerated
- Testing: 100% coverage — unit, property-based (fast-check), fuzz, and Playwright e2e for the web playground
- Build from scratch: Every pandas feature is implemented from first principles. No wrapping or porting existing JS/TS data libraries.
- Full feature parity with pandas — identical APIs adapted to TypeScript conventions and idiomatic structures. Every pandas feature is built from scratch, ground up, first principles. No ports of existing JS/TS data libraries.
- Interactive web playground — every feature ships with a rich, interactive tutorial, deployed to GitHub Pages. WASM where needed and useful.
- Performance — aggressive optimization throughout. Speed is a first-class concern.
- Exhaustive testing — pandas' own test suite as a baseline, extended with property-based testing, fuzzing, and e2e coverage. Target: 100%.