How about including the ids when you add to the db

chroma.add_texts(
    texts=split_texts, 
    embeddings=embeddings, 
    metadatas=[{'source': doc['source']} for doc in documents], 
    ids=ids  # Include the IDs here
)