Context

Creating an individual website has become cheap: a new project can be assembled in a day, and potentially this time can be reduced to hours. The bottleneck is shifting from developing a single website to managing multiple websites after launch.

With 10 websites, it is still tolerable to interact with each one individually. With 30–100 websites, this turns into constantly bypassing independent systems: to figure out where the required knowledge is located, an agent must first query a multitude of APIs; to modify a common element, one has to separately determine the state of each project.

The goal of this task is not to pre-determine a specific architecture, but to research what minimal common loop will allow an agent and a human to work with dozens and hundreds of websites as a managed set, without turning the websites into a single monolithic system.

What is already clear

• Each website must maintain its own independence and remain the source of truth for its own data and code.
• The agent needs a centralized way to find out what websites exist, what is on them, and what they are capable of, without prior traversal of each website.
• The information space of multiple websites is interconnected: new knowledge on one website may require revising materials on others.
• A similar problem arises with the software part: a common capability may need to be added or updated across dozens of projects at once.
• Growth in the number of websites should ideally translate primarily into growth in the volume of processed data, rather than a linear increase in manual organizational work.

Research directions

A potential direction appears to be a central informational and operational node that receives enough metadata from websites for search and coordination, while fetching full data and execution from the original website. This is currently a working hypothesis, not an approved architecture.

Sub-questions need to be researched separately:

• local website knowledge map;
• central information node;
• knowledge synchronization;
• search for connections and impact of changes;
• unified action management;
• mass software update;
• registry of websites and capabilities;
• connecting an arbitrary website;
• observability of a large cluster.

Connection to existing tasks

The loop must take into account already established haih-agent directions: distributed authorizations, addressable working memory for external AI clients, management of concept and fact quality, and executable task result checks. These address adjacent problems of agent identity, state, knowledge quality, and action verifiability, and must not be accidentally duplicated by the new layer.

Research result

The goal is to obtain a clear model where an agent is capable of working with a growing number of independent websites through a unified loop: quickly finding knowledge and affected websites, accessing the source of truth only when necessary, executing actions, and seeing the result. Specific protocols, data schemas, and technologies should be chosen after researching the sub-tasks.