An agent acts toward a goal; a browser agent does so on websites; an agentic browser builds that into the browser itself; human in the loop keeps a person in charge of final decisions.
A – C
Agent. Software that pursues a goal by choosing and performing actions over several steps. See What is an AI agent?
Agentic browser / AI browser. A web browser with an agent built in, such as Comet or Opera Neon.
Agent mode. A product setting in which the assistant may take actions rather than only reply.
Approval prompt. A request for the user to confirm an action before it happens.
Autonomy. How much an agent may do without asking.
Browser agent. An agent that operates websites: clicks, types, navigates.
Chatbot. A system that answers in text and takes no actions. See agent vs chatbot.
Computer use. An agent controlling a whole desktop through screenshots, mouse and keyboard.
Context. The information given to the model for a task: your message, the page, earlier turns.
D – M
DOM snapshot. A structured description of a page — its buttons, links, fields and text — that a browser agent reads instead of pixels.
Extension. An add-on for a browser. Hark is one.
Guardrails. Rules and checks that limit what an agent may do.
Hallucination. A confident but false statement by a model.
Human in the loop. A person approves consequential steps. See our guide.
LLM. Large language model, the reasoning engine inside chatbots and agents.
MCP. Model Context Protocol, an open standard for connecting AI systems to tools and data.
P – Z
Prompt. The instruction you give.
Prompt injection. Text in a page or document that tries to give an AI instructions the user did not intend. See agent safety.
Side panel. The column at the edge of Chrome where extensions like Hark live. See our guide.
Tool. An ability the model can call: click, type, search, send.
Tracker. In Hark, the list of orders, bookings, bills and deadlines saved from pages and emails.