cloro
Web Search API

The live web search API built for AI agents and RAG

Your agent's "search the web" tool is only as good as the results behind it. Bing's API is retired. cloro returns ranked results, snippets and sources as parsed JSON that your RAG pipeline or function-calling loop consumes directly. On SimpleQA Verified it scored 96.9% at $1.92 per 1,000 async searches.

4.7 · 66 reviewsG2.com software review platform logo
AI product teams grounding agents and RAG pipelines on cloro web search
5,000,000,000+ monthly API calls

What is a web search API?

A web search API lets your code run a web search and get the ranked results back as structured data: the programmatic version of typing a query into a search engine. Instead of HTML built for a browser, you get JSON built for a pipeline: ranked organic results with titles, URLs and snippets, plus the surrounding elements (ads, People Also Ask, related searches) when you want them.

Demand for web search APIs has shifted from search-engine tooling to AI applications: RAG pipelines and agents need live web results to ground their answers in current facts and citable sources. What separates providers is whether the results answer the question, and what each answer costs. The chart below measures both, and the web search API benchmark has the method and every result.

Web search API accuracy against cost

We ran three public question sets through six search APIs: one search per question, and the same model answers from each set of results. Cost is the list price per 1,000 searches on the plan that covers 100,000 searches a month.

Why AI developers pick cloro as their web search API

After Bing's API retirement, every web search API claims to be the one for agents. What matters is whether the results answer the question, and what each answer costs. We measured both.

Microsoft Lifecycle notice announcing Bing Search APIs retiring on August 11, 2025

The Bing Search API is gone: pick a replacement, not a clone

Microsoft retired the Bing Search API in August 2025 and stranded thousands of production integrations. The real choice now is which results sit behind your search tool. cloro returns ranked web results through one JSON endpoint, and they map onto the shape a Bing integration already reads. See our Bing Search API retirement guide for the migration path.

Accuracy you can check, at a price you can plan

Brave Search API, Tavily, Exa and Parallel are real products solving the same job, so we ran the same questions through all of them, with one model reading the results. On SimpleQA Verified, cloro scored 96.9%, within the margin of error of Parallel and Exa, at $1.92 per 1,000 async searches with the AI Overview against their $5 and $7. On FreshQA it scored 88.3%, ahead of Parallel, Exa and Tavily. On FRAMES, 824 multi-hop questions, it scored 80.2%, the highest of the six, while Parallel, Exa, Brave and Tavily scored 48.4% to 65.3%. A gap in results shows up as your agent answering "no results found", or hallucinating to fill the gap. The chart above has every result, and our web search API comparison adds price, free tiers and overlap with Google.

AI engine web search in action — the same grounding surface your agent gets from cloro

AI engines ground on web search, and your agent should too

When ChatGPT or Gemini answers a question, it fans the prompt out into web searches and synthesizes from the results (see our query fan-out analysis). Give your agent the same move and it answers with sources a user can click and check. Across roughly 2,500 prompts per engine, Google AI Mode grounded 98.2% of answers at 9.9 sources and AI Overview 91.2% at 9.1 (grounding study).

cloro pay-per-call pricing plans with per-engine credit rates — no subscription minimums

Searching for a "free web search API"? Do the math first

Most web search APIs price as subscription tiers with monthly minimums: the free tier is a trial, and the first paid tier charges you for capacity you may never use. cloro is pay-per-call: a search costs 3 credits, you start with free credits at signup, and you never pay for idle capacity. At Hobby ($100/month, 250k credits), that's ~83,000 searches. See pricing for the full math.

Use cases

Four workloads AI teams ship on top of cloro web search.

RAG grounding with citations

Fetch ranked results at question time, pack them into the context window, and return answers that cite verifiable source URLs.

Any LLMVector DB optionalcloro search
Best for: LLM app builders

Agent search tool

Register web_search as a function-calling tool; the model decides when to search and gets ranked, citable JSON back in the loop.

Function callingAgent frameworkcloro search
Best for: Agent developers

Fact-checking pipeline

Verify model outputs or user-submitted claims against live top-10 results before publishing — a search per claim, batched via async.

Async webhookClaim extractorcloro search
Best for: Trust & safety teams

Bing Search API migration

Swap the retired Bing endpoint for one POST call; map organicResults to your existing webPages.value shape and ship the same day.

One endpointThin adaptercloro search
Best for: Teams stranded by Bing's retirement

Wire live web search into your model in minutes

Send a query, cloro runs the search, ranked JSON comes back. Two patterns cover most AI workloads: fetch ranked results and pack them into your prompt (RAG grounding), or register cloro as a function-calling tool and let the model decide when to search.

Ground a RAG pipeline with live web results

python
from cloro import Cloro

client = Cloro(api_key="sk_live_your_api_key_here")

# 1. Search the web for grounding context.
response = client.monitor.google(
    query="eu ai act enforcement timeline",
    country="US",
    device="desktop",
)
results = response["result"]["organicResults"]

# 2. Pack ranked results into the model's context window.
context = "\n\n".join(
    f"[{r['position']}] {r['title']}\n{r['link']}\n{r['snippet']}"
    for r in results
)

prompt = f"""Answer using only the web results below.
Cite sources by [number].

Web results:
{context}

Question: When does EU AI Act enforcement start for GPAI models?"""

# 3. Send the prompt to your LLM of choice. Citations map back
#    to real ranked URLs your users can verify.
print(prompt)

Response example

200 OKapplication/json
{
  "success": true,
  "result": {
    "organicResults": [
      {
        "position": 1,
        "title": "EU AI Act: Implementation Timeline and Key Dates",
        "link": "https://artificialintelligenceact.eu/implementation-timeline/",
        "snippet": "Obligations for general-purpose AI models apply from August 2025, with full enforcement for high-risk systems phasing in through 2026–2027...",
        "page": 1
      },
      {
        "position": 2,
        "title": "AI Act enters into force — European Commission",
        "link": "https://commission.europa.eu/news/ai-act-enters-force",
        "snippet": "The AI Act's staggered application means providers of GPAI models placed on the market must comply with transparency requirements...",
        "page": 1
      }
    ],
    "peopleAlsoAsk": [
      {
        "question": "Is the EU AI Act already in force?",
        "type": "LINK",
        "snippet": "The AI Act entered into force in August 2024, but its obligations apply in stages..."
      }
    ],
    "relatedSearches": [
      {
        "query": "eu ai act gpai obligations"
      },
      {
        "query": "ai act high risk systems deadline"
      }
    ]
  }
}

Pricing that scales with you

Pick a plan that fits your volume. Price per credit drops as you scale.

Free
$0.00
per 1,000 credits
  • $0/mo
  • 500 credits
  • 1 concurrent job
  • 1 seat
  • Documentation assistant
Lite
$0.80
per 1,000 credits
  • $30/mo
  • 37,500 credits
  • 10 concurrent jobs
  • Unlimited seats
  • Email support
Hobby
$0.40
per 1,000 credits
  • $100/mo
  • 250,000 credits
  • 20 concurrent jobs
  • Unlimited seats
  • Email support
Starter
$0.39
per 1,000 credits
  • $250/mo
  • 650,000 credits
  • 50 concurrent jobs
  • Unlimited seats
  • Email support
Scale
$500+/mo
More credits
  • Volume discounts
  • More concurrency
  • Unlimited seats
  • Faster support
Compare all 11 tiers
PlanPrice / moCreditsPer 1kConcurrencySeatsSupport
Free$0500—11Documentation assistant
Lite$3037,500$0.8010UnlimitedEmail
Hobby$100250,000$0.4020UnlimitedEmail
Starter$250650,000$0.3950UnlimitedEmail
GrowthPopular$5001,350,000$0.3775UnlimitedPriority email
Business$1,0002,800,000$0.36100UnlimitedPriority email
Enterprise 2K$2,0005,871,025$0.34135UnlimitedSlack
Enterprise 3K$3,0009,306,606$0.32175UnlimitedSlack
Enterprise 4K$4,00012,756,261$0.31215UnlimitedSlack
Enterprise 5K$5,00016,391,783$0.31255UnlimitedSlack
Enterprise$5,000+Increased concurrency, overages on credits and credit discounts for annual contracts.Know more

Credit cost per request varies by provider. The figures below are for async/batch requests; sync requests add a +2 credit surcharge.

ChatGPT (full response) 7 credits
ChatGPT (web search) 5 credits
AI Mode 4 credits
Bing AI Mode 4 credits
AI Overview (incl. SERP) 5 credits
Gemini 4 credits
Google Search 3 credits +2/page
Google News 3 credits +2/page

ChatGPT full response includes query fan-out, ads, and shopping data. Google News uses the same pricing as Google Search.

Estimate your monthly cost and plan

7 credits each
5 credits each
4 credits each
4 credits each
4 credits each
3 credits / 1 page
3 credits / 1 page
Monthly requests
0
Credits needed
0
Recommended plan:

Web Search API, answered

What happened to the Bing Search API?

Microsoft retired the Bing Search API in August 2025, directing developers toward grounding inside Azure AI Agents instead of a raw search endpoint. That ended the only official general-web search API from a major engine. cloro fills the same slot: one POST call returns ranked web results as parsed JSON. Our Bing Search API guide walks through the retirement and migration in detail.

How does cloro compare to the Brave Search API?

Brave Search API queries Brave's own independent index, which is real engineering. We ran both on the same questions with the same reader model. On SimpleQA Verified (1,000 factual questions) cloro scored 96.9% against Brave's 93.1%. On FreshQA (358 questions whose answers change) cloro scored 88.3% against 84.9%, a gap inside the margin of error. At 100,000 searches a month, cloro costs $1.92 per 1,000 searches and Brave $5.

Can I use cloro as a search tool for my LLM agent?

Yes, this is the main use of the endpoint. Define a `web_search` function-calling tool in your agent framework (OpenAI tools, Anthropic tool use, LangChain and the Vercel AI SDK use the same pattern), and have it call `POST /v1/monitor/google`. Map `organicResults[]` to `{rank, title, url, snippet}` and return that array as the tool result. Our guide to adding web search to an AI agent has the Python code for grounding and for tool calling, and a TypeScript agent. Results arrive ranked and citable, so answer citations map to URLs your users can verify.

Is there a free web search API?

Fully-free general web search APIs don't survive. Running one against a real index costs real money, which is why free tiers are trials and "free" offerings get retired (see: Bing). cloro's honest version: free credits at signup to build and test your integration, then pay-per-call with no subscription minimum at 3 credits per search, so you pay for the searches your agent actually runs, not a monthly capacity guess. Details on the pricing page.

What does web search cost at production volume?

A search call is 3 credits (n=10 results); +2 credits per additional results page. The Hobby plan ($100/month, 250k credits) covers ~83,000 searches, enough for an agent product doing ~2,700 searches a day. Growth ($500/month, 1.35M credits) covers ~450,000 searches. If your pipeline also queries AI engines directly, those run 4 to 5 credits per call (ChatGPT web search 5, Gemini 4, AI Mode 4) on the same balance.

How do I handle burst traffic and concurrency?

Synchronous calls suit interactive agent loops (one user question, one or two searches, results in the response path). For batch workloads (nightly RAG index refresh, fact-checking a document queue), use `POST /v1/async/task` with `taskType: "GOOGLE"` and a webhook URL: submit the batch, results land at your webhook as each completes, and you never burn interactive concurrency on background jobs.

Do I get more than ranked links back?

Each response includes `organicResults[]` (position, title, link, snippet), `peopleAlsoAsk[]` (real user questions with answer snippets, useful for expanding agent research), and `relatedSearches[]` (query reformulations, useful for multi-hop retrieval). The full envelope is documented on the SERP API page.

What's the difference between this and cloro's SERP monitoring?

Same endpoint, different workload. SERP monitoring snapshots queries on a cadence to detect change over time — an SEO and competitive-intel job. A web search API is request-driven: your application searches when a user or agent needs fresh information, and the result is consumed immediately as grounding context, not warehoused as a time series. Both bill from the same credit balance with the same key.

Give your agent real web search today

Ranked web results as JSON, from your first free credits. 3 credits per search, pay only for what your agent runs. No subscription minimums, no index compromises.