Establish rights before ingestion
Define the sources you can access and reuse, the permitted purpose and retention. Public visibility does not automatically grant training or redistribution rights. Exclude private pages, credentials and unnecessary personal data from collection and embeddings.
Keep access and model work separate
The proxy carries network requests. Your application performs parsing, cleaning, chunking, embedding and retrieval. CloudCity proxy access does not include a model, vector database, structured extraction API or a license to the source content.
Build traceable chunks
Store canonical URL, retrieval time, source version or hash, extraction method and access decision alongside each document. Keep citations through chunking and retrieval. Deduplicate before embedding and test whether a retrieved answer still matches the current source.
Plan refresh and removal
Budget bandwidth for refreshes, failures and retries. Propagate source deletion or permission changes into the corpus and index. Evaluate retrieval quality with a known set of questions, and report missing or stale evidence rather than manufacturing a confident answer.