Browser-Harness is a coordinate/CDP-driven browser harness for AI agents. It is built for repeated browser work where the same navigation steps would otherwise be regenerated on every run, and its stated purpose is to keep token usage down by reusing a preconfigured navigation structure.
Its core behavior is simple: the site navigation is configured once, and the agent then reuses that structure instead of producing fresh Playwright scripts each time. The description of the tool emphasizes structural reuse, not caching resolved actions. In the comparison given on the page, it is said to sit between raw Playwright, which throws away generated scripts after each run, and Stagehand, which caches the resolved action outright. The tool is also associated with repeated browser flows, where avoiding rewritten navigation code keeps the token bill from climbing.
The page frames Browser-Harness in the context of browser automation for AI agents and mentions browser testing among its related topics. It is described as operating at the coordinate/CDP level rather than having the agent generate fresh Playwright scripts on every run. No pricing model, license, or packaging details are stated on the page beyond the article label “Free Token Saving.”
Browser-Harness is a Developer Tools project. It reduces redundant token usage and script regeneration for AI agents performing repeated browser automation tasks. Browser-Harness is an open-source project aimed at AI agent developers. Browser-Harness is free to use. It runs on the web and the command line.
Behind Browser-Harness is Steven Gonsalvez, and it first shipped in 2026. Development happens publicly on GitHub with 15.9k stars and 442 commits in the last 90 days. Among its 5 catalogued features are preconfigured navigation, token optimization, and CDP integration.
Summary written by a language model from the project’s public pages.
What PulseGate has recorded for this listing
Closest matches by what these projects do