Skip to content

Bajutsu

Natural-language-driven, determinism-first E2E testing on a backend-agnostic driver — one scenario format and one deterministic runner, where a platform is just a backend. AI authors and investigates; a machine decides pass or fail.

Home

Bajutsu (馬術) takes test scenarios written in, or recorded from, natural language, drives your app (taps, typing, swipes, waits), and verifies the result with machine-checkable assertions. Everything but one seam is platform-neutral. That seam is the backend that actuates the UI, so swapping the backend runs the same scenario, unchanged, on a different target.

The core stance

  • AI is the author and the failure investigator, never the judge. AI helps write scenarios and investigate failures, but a run is fully deterministic — pass/fail comes only from machine assertions, never a model.
  • Two tiers. Tier 1 is AI live operation (exploration and authoring); Tier 2 is the deterministic runner that gates CI.
  • Determinism first. Condition waits only (the one fixed pause is an explicit, capped sleep step that must state its reason). An ambiguous selector fails immediately instead of tapping whatever matched first.
  • A platform is a backend. The deterministic core names no platform; add or swap a backend and the same scenario format, runner, and CLI target a new platform unchanged.

Status — pre-alpha

The deterministic core, the AI authoring loop (record / crawl), the evidence subsystem, codegen, and self-healing triage are all implemented and unit-tested, and those unit tests need no Simulator. The iOS XCUITest backend is validated end-to-end on a real Simulator, the Android (adb) backend is validated on an emulator, and the web (Playwright) backend runs a deterministic run against a browser on the Linux gate. Flutter apps run through those same backends unchanged, with no separate Flutter backend.

Quickstart

uv sync --group dev                                       # .venv + deps + dev tools
bajutsu record --target <name> --goal "..." [--out f.yaml]  # Tier 1: AI explore + record
bajutsu run    --target <name> [--scenario f.yaml]          # Tier 2: deterministic pass/fail

The Getting started tutorial walks through install, unit tests, scenario authoring, a device run, and the report. On a machine without a Mac, the web track runs the same loop against a browser (Playwright backend), with no Xcode or Simulator.

Feature highlights

  • Author with AI — record and crawl explore the app and write scenarios you can hand-edit.
  • Deterministic runner — condition waits, a stability ladder, and selectors that fail loudly on ambiguity.
  • Evidence subsystem — screenshots, hierarchy, and network captured on a policy you control.
  • Self-healing triage — AI investigates a failure and proposes a fix, off the CI path.
  • Codegen — turn a scenario into an equivalent native test: XCUITest (iOS), Playwright (web), or UI Automator (Android).
  • MCP & web UI — drive Bajutsu from an MCP client or the local serve UI.

Backends & platforms

Platform Backend Status
iOS Simulator XCUITest Validated end-to-end on-device
Web Playwright Validated end-to-end on the Linux gate
Android adb Validated end-to-end on an emulator
Flutter via the XCUITest / adb backends Validated end-to-end on-device

Learn more