Kimi’s Browser Extension Turns Web Actions into Reusable Skills

Introduction

As AI agents move beyond answering questions, the browser is becoming an execution surface rather than just a place to display information. Kimi has introduced a browser extension based on its earlier Kimi WebBridge, adding a sidebar interface and a way to record web actions as reusable Skills.

What has changed

The extension keeps WebBridge’s main capabilities. With the user’s authorization, Kimi can use Chrome or Edge to open pages, read their contents, click controls, and carry out browser-based tasks. The earlier connection method, in which a local agent called the extension, remains available. At the same time, users can now click the Kimi icon in the browser toolbar and start a conversation directly from a sidebar.

The more meaningful change is the addition of workflow recording. Instead of asking the agent to rediscover the same procedure every time, a user can record a sequence of actions and turn it into a Skill. A routine such as opening a particular website, finding the day’s information, and exporting data can therefore be saved for future use. Kimi can also save a successful session in Skill form.

Why it matters

Browser agents have traditionally required repeated instruction and adjustment. A changed page layout, a relocated button, or dynamically loaded content can easily disrupt execution. Recording does not remove these reliability problems, but it lowers the cost of describing repetitive procedures and gives the agent a way to retain a practical workflow.

This helps explain why browser extensions remain relevant. Earlier AI extensions focused mainly on reading: answering questions beside a page, translating foreign-language content, filtering videos, or assisting with papers. The newer direction is “using the web” on the user’s behalf. Kimi’s extension follows that shift by combining page understanding with direct browser interaction and workflow reuse.

Connection with Kimi Code Desktop

Kimi Code Desktop targets software projects and includes a browser for checking pages generated by an agent. The browser extension serves a different setting: websites that users already access through Chrome or Edge. One connects the agent to project files, terminals, and development history; the other connects it to the user’s external web environment.

Together, the two products suggest a broader agent strategy. Kimi can work inside a development workspace, inspect the result in an embedded browser, and then reach everyday web services through the user’s existing browser. The boundary between desktop software and online operations becomes more flexible.

The limitations are still clear. Website redesigns, dynamic loading, and complicated interactions may cause a task to fail. Users may need to inspect the page, confirm its current state, and refine their instructions. The main significance of this release is therefore not simply another chat entry point. It is the attempt to turn one-off browser actions into repeatable procedures that can gradually become agent Skills.

Source: QbitAI