5Qj*ddlmZmZGddeZy))ABCabstractmethodc4eZdZdZededeedefdZy) LLMProvideraProvider-agnostic interface every LLM backend must implement. Every service (itinerary generation, reply drafting, ...) is written against this interface only - never against a vendor SDK directly. Swapping OpenAI for Anthropic, or cloud for a local Ollama model, is a one-line change in providers/factory.py and an env var; nothing in app/services/ or app/prompts.py has to change. system_promptmessagesreturnct)aDRun one completion and return the parsed JSON object the model produced. `messages` is a list of {"role": "user"|"assistant", "content": str} - the system prompt is passed separately since not every provider's wire format allows a "system" entry inside the messages list (Anthropic doesn't). Raises whatever the underlying SDK raises on a transport/API error, and json.JSONDecodeError if the model's output can't be parsed as JSON even after the implementation's own cleanup - callers are expected to catch and wrap both. )NotImplementedError)selfrrs 8/var/www/html/meridian/meridian-ai/app/providers/base.py complete_jsonzLLMProvider.complete_json s "!N) __name__ __module__ __qualname____doc__rstrlistdictrrr rrs4 "3 "$t* " " "rrN)abcrrrrrr rs#"#"r