I’ve noticed that when I ask the RTILA assistant to build a project, it sometimes magically remembers a custom skill or a specific JSON schema I used in a completely different project last week. I didn’t put it in my prompt.
Is it using RAG (Retrieval-Augmented Generation) under the hood? How does it decide what context to pull?
Good eye! Yes, RTILA X features a fully integrated, local RAG pipeline to give your AI long-term memory.
We embedded a local Qdrant vector database directly into the app. When you type a prompt, we use a local embedding model to instantly convert your query into a vector and search your past projects, successful scripts, and saved skills.
If it finds highly relevant matches (based on the similarity threshold you set in Preferences), it silently injects those past examples into the LLM’s system prompt before generating the response. This means the more you use RTILA, the smarter the AI gets at matching your specific coding style and target websites - all without sending your history to a third-party server.