LaVague

LaVague is a developer-focused framework that aims to reduce the pain of maintaining web automation scripts. Instead of hand-coding selectors and waiting for pages to change, you describe an objective and the framework generates and executes the steps. The World Model and Action Engine split the problem: one handles reasoning, the other handles code generation and execution. This separation is clean and extensible. Key strengths: The abstraction level is high—you can automate multi-step workflows without touching low-level selectors. Support for multiple drivers (Selenium, Playwright, Chrome Extension) gives flexibility. The TokenCounter helps you monitor costs, which is critical when every run burns tokens. LaVague QA turns Gherkin specs into tests, a boon for QA teams already using behavior-driven development. Weaknesses: You need Python and API keys, which rules out non-programmers. The framework is young, so breaking changes are possible. Performance depends on the LLM; complex or multi-page objectives might fail. The project's roadmap and community are active, but it's not a mature commercial product with enterprise support. Where it fits: Internal tools, prototyping, and repetitive web tasks where you'd otherwise write brittle scripts. QA teams can extend their test suites with LaVague QA. Where it doesn't: large-scale production deployments without robust testing, and teams without Python skills. Alternatives: SikuliX, AutoHotkey, or UiPath for visual automation; but they don't offer the same intelligence. Hand-coded Playwright/Puppeteer gives more control but requires more maintenance.