Retrieval-Augmented Generation using Remember as a document store and Recall as the retriever. ConditionalRouter catches LLM errors gracefully, splitting output into success and error paths.
Store docs → search → build RAG prompt → LLM answers → route errors vs. success.
Remember stores documents as plain text files with keyword-searchable titles. No vector database needed. Recall retrieves them by keyword match — fast and transparent.
ConditionalRouter checks for "LLM_ERROR" in the response. If the LLM is down or returns an error, it goes to the error display. Success goes to the answer display. No broken pipes.
The PromptTemplate enforces "answer using ONLY the provided context." The LLM is set to low temperature (0.3) for factual responses. If context doesn't cover the question, it says so.
Add more Remember nodes to ingest multiple documents. Add more Recall nodes with different queries to search different aspects. Combine results with TextCombine before the LLM.