Give your AI agents live web access
A web search API for AI agents that plugs straight into your existing tool stack. Autonomous agents call /search, /crawl, and /research as tools to ground their reasoning in current information. Same Tavily endpoints and response shapes, one-line migration, half the per-credit cost.
/grounding
Agents that can’t search are guessing
A model alone is frozen at its training cutoff. The moment an agent needs today’s price, a fresh release note, or a fact it never saw, it needs a way to reach the open web. A search tool turns a static model into an agent that can verify, gather, and act on current information.
Grounded answers
Cut hallucinations by letting the agent cite live sources instead of recalling stale facts.
Autonomous loops
Search, read the result, decide, search again. The agent drives its own research until the task is done.
Real freshness
From a quick lookup to a deep multi-step dive, the same agent reaches whatever the web knows right now.
/tools
Five endpoints, five agent tools
Register each endpoint as a tool the model can call. Give it a clear name and description, point the handler at the endpoint, and hand the JSON back to the model. These are the exact Tavily endpoints, so existing tool definitions carry over unchanged.
The agent’s default research move. Return ranked results with snippets the model can reason over before it acts.
Pull clean, readable content from a known URL so the agent ingests the article, not the navigation chrome.
Follow links from a starting point to gather documentation, product pages, or knowledge across a domain.
Discover a site’s URL graph first, so the agent plans what to crawl instead of fetching blindly.
Multi-step research in one call — ideal as a single high-leverage tool for deep-dive agent workflows.
/migration
Your agent framework already supports it
Because every request and response matches Tavily 1:1, any agent framework that can already call Tavily works with BestSearch the instant you repoint the base URL. No new SDK, no schema rewrites, no changes to how the model reads the tool output.
Define the search tool
# Point your agent's search tool at BestSearch
BASE_URL = "https://app.websearchapi.tech"
def web_search(query: str) -> dict:
return http_post(
BASE_URL + "/search",
json={"query": query, "max_results": 5},
headers={"Authorization": "Bearer " + API_KEY},
).json() # same shape as TavilyRegister it for function calling
tool = {
"name": "web_search",
"description": "Search the live web",
"input_schema": {
"type": "object",
"properties": {
"query": {"type": "string"}
},
"required": ["query"],
},
}
# OpenAI- or Anthropic-style tool use:
# the model calls web_search, you return JSON.Already on Tavily? Migration is one line: swap the base URL to https://app.websearchapi.tech. Everything else stays the same.
/economics
Half price, where it compounds
Agent workloads burn credits in loops, not single calls. A deep-research run can chain dozens of searches and reads. The same credit model as Tavily means the count is identical — you just pay half per credit.
Tavily
$0.008 / credit
List per-credit price. Every agent loop multiplies it.
BestSearch
$0.004 / credit
Same credits, same behavior — exactly half the cost at agent scale.
/faq
Questions from agent builders
Does my existing Tavily-based agent framework work with BestSearch?
Yes. Any framework or agent that already calls Tavily works by repointing the base URL to https://app.websearchapi.tech. The endpoints (/search, /extract, /crawl, /map, /research), request parameters, and response shapes are identical, so your tool definitions and parsing code stay exactly as they are.
How do I expose web search as a tool for function calling?
Define a tool whose handler issues an HTTP POST to /search (or /crawl, /research) and returns the JSON to the model. Because the response shape matches Tavily, any OpenAI- or Anthropic-style tool/function-calling setup that already parses Tavily output needs no changes — only the base URL differs.
Why does price matter for agents specifically?
Autonomous agents loop — they search, read, reason, and search again, often many times per task. That multiplies credit usage fast. At $0.004 per credit versus Tavily’s $0.008, the same agent behavior costs half as much, with no change to capability.
Will switching change my agent’s behavior or output?
No. BestSearch mirrors Tavily’s endpoints, parameters, response fields, and credit model. Migration is a one-line base-URL change, so your agent receives the same data structures and behaves the same — it just costs less per run.
Ship agents that can reach the web
Wire up the search tool, repoint the base URL, and let your agents loop. Tavily-compatible, half the price.