AgenticME

Browser agents are experiencing a surge in popularity due to their ability to bridge the gap between large language models and web functionality. However, current solutions operate in the cloud with no absolute guarantees about users' data and account credentials entrusted to these browser agents (e.g. OperatorAI, Manus.im, Claude Computer Use). Leaving them exposed to potential hacks.

To be able to fully trust our browser agents with sensitive data, we explore an open-source solution called browser-use and deploy it in a TEE. This ensures that the input data, account credentials and the whole agent's task ( ! ) stay hidden from the cloud provider and completely safe to potential hacks.

We are using browser-use with connection to OpenAI to perform agentic tasks (this could in the future be upgraded to a locally running LLM). We use a web interface to schedule and monitor the tasks. The whole system is to be deployed in a TEE (such as Marlin Oyster Enclave).