AI-powered Intelligent Controllers
Upbound Crossplane supports running Intelligent Controllers, which are a class of composition functions and operation functions that bring LLM-driven logic into your control plane's reconcile loop.
Intelligent controllers extend the traditional observe → compare → act pattern
to observe → analyze → act → adapt cycles that incorporate AI-driven reasoning
capabilities. The analyze phase introduces decision making that can balance
competing objectives, understand business context, and optimize based on complex
patterns that deterministic logic can't easily capture. The adapt phase enables
learning from operational outcomes to improve future decision-making processes.
Intelligent Compositions
When you incorporate a composition function having AI intelligence to a composition pipeline, it's considered an intelligent composition. Some examples of these functions are:
These functions pass the pipeline context to a Large Language Model, such as Claude, and provide user-specified prompts to task Claude with influencing the configuration of a function pipeline's desired resources. MCP servers, packaged as Add-Ons, are installable on your control plane and deliver additional tools that may be used by the language model.
Intelligent Operations
When you incorporate an operation function having AI intelligence to an operation pipeline, it's considered an intelligent operation. Some examples of these functions are:
- [function-ai-status-transformer][function-ai-status-transformer]
- [function-pod-analyzer][function-pod-analyzer]
Model Context Protocol (MCP) Servers
Model Context Protocol (MCP) is an open standard for connecting LLMs to systems where data lives. Upbound Crossplane supports the installation of MCP servers, packaged as Add-Ons. Some examples are:
Learn about how to package an MCP server into an Add-On in the concept documentation.
Next steps
Learn about example use cases of Intelligent Controllers by reading the guides