Real-time web search for your LLM apps

/grounding

Why ground an LLM in live web results

A language model only knows what it was trained on. Retrieval-augmented generation closes that gap: you fetch real, current pages at query time and let the model answer from text it can actually quote.

/endpoints

Two endpoints do the grounding work

Most LLM grounding flows need just two calls. Both match Tavily one-to-one in parameters and response shape.

/example

Grounding in a few lines

Search the live web, then hand the results to your model as context. Migrating from Tavily? Change one line — the base URL.

# One-line switch: point Tavily's base URL at BestSearch
import requests

BASE = "https://api.websearchapi.tech"   # was: api.tavily.com
KEY  = "bs-..."

# 1. Fresh web results to ground the answer
results = requests.post(f"{BASE}/search", json={
    "api_key": KEY,
    "query": "latest changes to the OpenAI rate limits",
    "search_depth": "advanced",
    "max_results": 5,
}).json()

# 2. Build grounding context + citations from the results
context = "\n\n".join(
    f"[{r['title']}]({r['url']})\n{r['content']}"
    for r in results["results"]
)

# 3. Pass context into your LLM prompt
prompt = (
    "Answer using ONLY the sources below. Cite each claim.\n\n"
    f"Sources:\n{context}\n\nQuestion: What changed?"
)
# -> send prompt to your model; it answers from real, cited text

/price

Same grounding, half the bill

Grounded LLM apps call search on every turn, so cost compounds fast. BestSearch keeps the Tavily credit model and simply halves the per-credit price.

/faq

Frequently asked questions

How does a web search API reduce LLM hallucination?

Instead of relying on stale training data, your model retrieves fresh sources at query time. You pass /search results into the prompt as grounding context, and the LLM answers from real text it can quote — so you can attach citations and the model is far less likely to invent facts.

Is this really Tavily-compatible?

Yes. BestSearch exposes the same /search and /extract endpoints with the same request parameters and response shapes Tavily uses. If your RAG or agent code already calls Tavily, you repoint the base URL and keep your existing SDK code and prompts.

When should I use /search versus /extract?

Use /search to get fresh, ranked results with snippets for a query — ideal for grounding a quick answer. Use /extract when you already have a URL and need the full clean page content, for example to feed a long source into a summarization or analysis step.

How much does it cost compared to Tavily?

BestSearch uses the same Tavily credit model but charges $0.004 per credit versus Tavily's $0.008 — exactly half. A grounded answer that cost you a dollar on Tavily costs fifty cents here, with no change to your code.