In this talk, Sergej Zuev presents a blueprint for an open-source AI infrastructure that integrates Plone as one of many tools in a composable, pluggable system. Built around a vectorstore-based data layer, self-hosted or open LLMs, and tool orchestration via MCP (Model Context Protocol), the setup allows flexible chaining of capabilities – from content generation to semantic search and automation.

Key components include:

• Pluggable LLMs (e.g. Ollama, LM Studio, GPT4All) with secure, local control
• MCP for orchestrating AI tools, including Plone, via API and container logic
• n8n for workflow automation and connecting external services or databases
• Use cases such as AI-assisted knowledge retrieval, content workflows, and chat-based editorial tools

This talk is ideal for developers, architects, and tech leads looking to build sustainable AI systems without vendor lock-in – using open tools to build custom intelligence, step by step. Takeaways: A real-world-ready infrastructure stack, lessons from building hybrid systems, and a vision of CMS as one node in a larger AI ecosystem