Browser agents need a policy for choosing the next action as a page changes. The examples here put Jev between a goal or command and existing browser tools: the surrounding application reads the page, offers actions, and performs the chosen step. This collection focuses on navigation and interaction, rather than grouping every Chrome extension under the same claim.
New to the model? What is Jev AI explains its state-to-decision interface.
Browse Jev case topics
A browser loop with explicit actions
Read the current page and task context through the browser adapter, DOM, or another available observation tool.
Present the actions available at this step, such as following a link, returning to a page, or selecting a visible control.
Execute the selected action through the browser tool and check the resulting page before continuing.
What the browser adapter still owns
Page observation, tool execution, permissions, and completion checks remain application responsibilities. A fast action choice does not by itself establish that a task finished correctly. Compare the actual task, permitted actions, fallback behavior, and success evidence in each source. The original authors’ timing and cost figures are not a shared JevPlay benchmark.
Selected community demos
Each selection links to its original author’s post. Source text and media come from the existing Jev community archive.
A job crawler that chooses links and roles
Kai starts from a company homepage, finds the Careers entry point, follows relevant links, and scores job matches. The demo illustrates repeated decisions across websites with different navigation structures.
Christopher shares a demo of automatically declining cookie banners and popups on several websites. It is a narrow browser task with a specific desired action.
Moritz Kremb sends a spoken command’s transcript to Jev and uses the returned probabilities to choose a browser action. The source video includes a request to go back.
Vlad Terin presents an open-source adapter for existing Codex browser tools. The post separates planning by Codex, action selection by Jev, and execution by the browser.
Gregor’s flight-search demo uses DOM state and a new action space at each step, with an LLM fallback for typing. The original post links a small open-source browser agent.
@princecaarlo connects agent-browser to a Jev loop with the goal of navigating from Coffee to Artificial Intelligence. The task provides a concrete destination for judging the navigation outcome.
Explore a smaller decision task: interpret a message and inspect its suggested intent and next action.
JevPlay is an independent project and is not affiliated with TypeSafe. Community posts, media, and trademarks belong to their original owners. Browse the full Jev case library for more sources.
Kai@hqmank
