Skip to main content

​
The problem

Support teams field the same questions over and over. Generic chatbots make it worse. They hallucinate answers that aren’t in your policies, creating wrong expectations and more tickets.

​
How Ragrails solves it

Ground the bot in your real help content, and make it honest when it doesn’t know.
  1. Ingest your help center and policy docs into one collection.
  2. Chat with retrieval-quality guardrails so weak matches get a careful answer or a refusal, never a hallucination.
  3. Rewrite follow-up questions so multi-turn conversations stay on track.
from ragrails import RagRails, ChatRetrievalQualityConfig, QueryRewriteConfig

rag = RagRails()
rag.setup_url()

# Build the support knowledge base
rag.ingest(
    urls="https://help.example.com",            # scrape the help center
    docs={"folder": "files/policies/"},          # parse policy PDFs
    embedding={"provider": "voyage", "model": "voyage-3"},
    storage={"vector_db": "qdrant", "collection": "support", "url": "http://localhost:6333"},
)

# Answer a ticket, refusing rather than guessing on weak matches
llm = rag.llm(provider="openai", model="gpt-4o-mini")
embedder = rag.embedder(provider="voyage", model="voyage-3", input_type="query")

result = rag.chat(
    "Can I get a refund after 30 days?",
    llm=llm, embedder=embedder,
    vector_db="qdrant", collection="support", url="http://localhost:6333",
    history=[],
    query_rewrite=QueryRewriteConfig(enabled=True),
    retrieval_quality=ChatRetrievalQualityConfig(low_confidence_mode="refuse_grounded_answer"),
)
print(result.answer, result.answer_confidence)

​
Features used

Ingestion · Chat · Answer quality · Query rewriting