Remember the visions of the future where smart services on the Internet would know everything about you and advice you real time and do things for you automatically?! With AI agents that vision seems closer than ever. But do you really want any kind of AI agent out there get access to all sorts of information about you?! And maybe control aspects of your life that you have full control of today?
ActingWeb enables communications between AI agents acting on behalf of humans and supports controlled sharing of information, including personal and sensitive information. It is on open protocol backed by a production-grade python SDK that allows humans to control how agents share data and how they act on their behalf. The MCP protocol enables AI agents to get access to tools from 3rd parties and the A2A protocol enables agents to collaborate on tasks and act like one. But you don’t want your insurance company’s AI to get access to all your personal health data, only your trusted (AI) doctor should get that. ActingWeb sits in between AI agents and groups of AI agents and allows humans to control what flows between them.
More in-depth about the technology.
The ActingWeb specification has all the details of the REST protocol needed for the app to app communication, as well as the application model that enables the extreme focus on a specific user or “thing”. A recent update has added full MCP server support, thus allowing AI services to connect directly into an ActingWeb mini-app.
The specification is backed by working code in the form of a reference implementation in python. The reference implementation is at all times running at https://demo.actingweb.io (you can even use it to test your own apps). You can also see ActingWeb working live in the AI Personal Memory application that offers a privacy-centric storage for all sorts of personal information you may want to share with AI chat or AI agent applications (over MCP).
RSS - Posts