Introducing CORE

Summary

The video introduces Site Star, an AI-assisted website platform, and Core, the reusable application stack that powers it. Greg describes both as products built from years of web development and client work, with an emphasis on automation, consistency, and structured engineering rather than letting AI generate an uncontrolled codebase.

Site Star: Managing a Website Through an Agent

Site Star is initially aimed at small businesses that find website maintenance expensive or time-consuming. Instead of relying on a developer for routine updates, a business owner can interact with their site through a dashboard, email, text message, phone call, or an AI assistant connected over MCP. These channels feed into the same backend and agent workflow.

The demonstration shows users asking questions about their site—such as visitor numbers, bot activity, or SEO information—or requesting changes. In Greg’s example, he calls the site and asks it to change a homepage button, then receives a later message when the change is ready or published. Requests can also prompt follow-up questions or require approval.

Site Star can connect to services such as Google Search Console and SEMrush. It also supports scheduled reports and automations, such as sending traffic information every week. Its analytics include server-side information, including which pages search or AI bots visit, in addition to ordinary website activity.

Site Star’s Background Agents

Greg describes a group of roughly 10–12 agents that periodically inspect a site from different perspectives, including SEO, accuracy, spelling, and errors. They share observations through a database-backed message board and may surface recommendations as “ideas” for the owner to approve. One demonstration suggestion flags an overly long page title.

The platform is designed to monitor and respond to problems as well. Errors and operational signals can enter automated workflows, with agents potentially investigating or fixing issues. Agents have different roles and tool permissions, so they can access only the capabilities allowed for their tasks.

Messages and interactions from different channels are stored in a conversation database. Greg says the system can vectorize content and support semantic and lexical search, allowing agents to retrieve relevant history and website context.

Core: A Configuration-Driven Application Stack

Core is Greg’s reusable foundation for building web applications. It includes common infrastructure such as registration and authentication, authorization, user roles and scopes, APIs, API keys and usage tracking, data models, and MCP surfaces.

Much of an application is configured through a config.json file, which controls features such as exposed APIs, permissions, roles, scopes, and other application settings. Greg says a new deployment can go from initial setup to roughly 60% of an application in about ten minutes, allowing development to focus sooner on the application’s business logic. Core can be deployed on AWS and configured through code or, in some cases, through an interface without directly changing the code.

Core supports both dedicated deployments and shared, multi-tenant deployments. Site Star’s platform and backend are separate Core deployments, while the company CRM and affiliate portal are also built on Core. Shared Core updates can be brought into applications through a versioned Git submodule, allowing fixes and improvements to be reused across products.

Greg says Core began as an internal foundation and has not yet been broadly released as a standalone product. He is open to giving interested people an opportunity to see it but does not want to open registration widely yet.

Other Developer Tools and Examples

• CRM and affiliate portal: The CRM supports sales operations, phone calls, marketing materials, and AI-assisted prospect research. The affiliate portal supports referral links, company information, and branded materials. Both expose APIs and MCP access, with permissions based on the user.
• LLM EXE: Greg’s TypeScript components for building LLM applications, originally developed around 2022–2023. The package separates prompts, models, and output parsing, and abstracts differences between providers. Greg says Site Star primarily uses Anthropic, with fallbacks to other providers.
• Conversation database: A storage and retrieval layer used for agent messages, semantic search, and conversation context.

Review, Analysis, and Demonstrations

The host gives a strongly positive assessment of Greg’s engineering approach, especially its focus on reliability, logging, error handling, security, and reusable infrastructure. The host contrasts Core’s constrained, template-based AI development with “vibe coding”: the AI is expected to work within established patterns rather than inventing an application structure from scratch. This is an enthusiastic endorsement, not a formal independent product review.

The video includes demonstrations of connecting an AI assistant to Site Star through MCP, requesting site information and changes, setting up automations, and viewing the site’s agent-generated ideas and analytics. It also shows examples of Core-powered products, including the CRM and affiliate portal.

Main Speakers and Sources

• Greg Reindel (whose surname appears as “Rindahl” in the subtitles): developer and creator of Site Star, Core, and related tools.
• Host of Unsupervised Learning: interviewer and commentator; the subtitles do not provide the host’s name.