A shared browser for humans and agents

A lot of work still happens in a browser. We research suppliers, update records, compare products, fill out forms, and move information between applications. These tasks often mix repetitive steps with moments that need a person’s judgment.

As agents take on more of that work, the browser becomes a place where people and software need to meet. An agent might prepare a task. A person might resolve a login challenge, correct a misunderstanding, or make a decision. Then the agent needs to continue from where the person left off.

BrowserFabric’s vision is to give them a shared place to do that: a remote Chrome browser built for both humans and AI agents. Today, we run Chromium in the cloud and provide the tools to operate it programmatically and interact with it through a live viewer. We want that shared browser to become a natural part of working with an agent.

The browser is where the work lives

When you hand a task to an agent, it needs more than an instruction. It needs somewhere to open pages, use applications, and keep track of what has changed.

A cloud browser gives the task a place to run outside an individual laptop. The agent can reach it from its own environment. A person can open a live view of the session. Both can work with the actual page, including the form that is halfway filled in or the result that needs checking.

Keeping that context together matters. A handoff should preserve the work already done. You should be able to step into the browser, understand the situation, and help without reconstructing the task from screenshots and a long chat history.

A handoff should be an ordinary part of the task

Imagine asking an agent to prepare a request for quotes from a supplier. It finds the right page and fills in the information you provided. The site asks for a login confirmation. You open the same browser, complete that step, and check the details. The agent can then inspect the updated page and continue under your instructions.

That is the kind of collaboration we want BrowserFabric to make easy: useful automation with a clear way for a person to participate.

Today, BrowserFabric supplies the live browser and human-input tools that a developer can use to build this workflow. The application still needs to coordinate when its agent pauses and resumes. After a human changes the page, the agent should observe it again before acting. Clear control ownership and dependable handoffs are central to the experience we want to build.

Agents need a browser they can understand

A useful agent needs to know what is on the page, what it can interact with, and whether its last action worked. That is why BrowserFabric’s agent interface follows a simple sequence: observe, act, and verify.

The agent observes the current page, acts on the controls it finds, and checks the result. When the page changes or someone intervenes, it can take a fresh look. This gives developers a practical foundation for building workflows that respond to the browser’s actual state.

We want developers to be able to choose the agent that fits their work. BrowserFabric can be used from Codex, Claude, or a custom cloud agent through instructions and our documented interfaces. REST, MCP, and SDK access let the browser fit into an existing workflow.

Humans need visibility and a way to intervene

For a person, the important questions are straightforward: What is the agent doing? Is it working on the right page? Can I help when it gets stuck? Can I review the result before the next step?

The live viewer makes the remote browser visible and provides a way to interact with it. Our broader goal is to make participation feel natural: watch progress, step in when your judgment is needed, and return the task to the agent with clear context.

That also means making control understandable. Who is acting, what has changed, and which decisions need a person should be easy to see. We see these as core design responsibilities for a shared browser.

Build on the session, then on the relationship

BrowserFabric already brings together cloud Chromium, agent tools, a live viewer with human input, and saved browser contexts. Saved cookies and browser storage can help a later session pick up with useful login state, subject to the website’s own session rules.

Our direction is to build a more complete collaborative experience on that foundation: clearer handoffs, better visibility into progress, and continuity when a task changes hands. These are the things we want to make easier as the product develops.

The browser is a familiar place for people and an increasingly capable place for agents. We believe it can become a shared workspace for both. BrowserFabric is our effort to make that workspace accessible from wherever you and your agents run.