RAG retrieval — feed your LLM live web results

Live web retrieval as your RAG layer — citable context in one request.

Read the search API docs

RAG quality hinges on the retrieval layer’s quality and freshness. TutuFind provides real-time web retrieval: one request returns structured results with explicit sources to drop straight into model context.

Results carry source attribution for citing provenance; llms.txt is supported and Chinese is jieba-tokenized — covering both English and Chinese RAG.

Funciones principales

Real-time & fresh

Retrieve the latest web content to fill knowledge gaps past the model’s training cutoff.

Citable

Each result carries source and url for cited answers and lower hallucination.

Clean context

Structured title/snippet reduce cleanup before feeding the model.

EN & CN

jieba tokenization + web fallback recall long-tail questions too.

Cómo se usa

  1. 1
    Take the question
    Pass the user query to TutuFind.
  2. 2
    Retrieve top-k
    GET /api/v1/search?q=…, take the first few snippet / url / source.
  3. 3
    Assemble the prompt
    Inject retrieved snippets and sources into your context template.
  4. 4
    Generate cited answer
    Have the LLM answer from retrieved content and cite sources.

Preguntas frecuentes

Why use TutuFind as the RAG retrieval layer?
One request yields structured, traceable, real-time results — no self-built crawler / parser / anti-bot before assembling context.
How to reduce hallucination?
Each result has url and source; instruct the model to answer only from retrieved snippets and cite them.
Does it support Chinese documents?
Yes, jieba tokenization improves Chinese recall.
How fresh are results?
The local index updates continuously via the crawler, with web fallback filling in fresh content.

Integra ya la API de TutuFind

Cuota gratuita · resultados estructurados con una sola petición · compatible con llms.txt

Read the search API docs