An LLM connection lets your Python services call Claude, OpenAI, Gemini or any self-hosted model behind one API. The provider's credentials stay in the platform, every call leaves an audit trail, and chat history, skills and GitOps come built in.
Call a model from a service with self.llm, read the reply and the per-call token usage
Conversations with managed history - the platform keeps each chat's turns and trims what goes to the provider
Reusable instructions sent as the system context of a call, shared with MCP gateways
Providers and models, addresses, API keys, pool size, timeout, token and history limits
Export and import the whole AI configuration as YAML across dev, test and production
Every call is recorded with its outcome and duration, feeding the built-in alerting
A one-on-one session covers LLM connections and MCP gateways end to end.