Guide

Free web search API for AI agents

An LLM knows only its training data. A web search API gives it current facts and sources. Tavily, Exa, Firecrawl and Linkup each give free monthly credits. FreeLLMAPI puts these engines behind one /v1/search endpoint and moves to the next engine when one fails.

Free search engines today

The table lists the search engines in the FreeLLMAPI catalog. The data comes from the live catalog. "Answers" means that the engine can return a short written answer with the results. "News" means that it can limit the search to news.

Search APIFree limitAnswersNews
Tavily Search1,000 credits/month, no cardYesYes
Exa Search$10/month in credits, no cardNoYes
Firecrawl Search1,000 credits/month (2 per search), no cardNoYes
Linkup Searchfree monthly credits for eligible accountsYesNo
Parallel SearchFree, no key needed (rate-limited); free API key for moreNoNo
SerpSearch1,000 requests/month shared across endpoints; no card; 1 request/secondNoYes

Live data from the FreeLLMAPI catalog (version 2026.11.19, checked 19 November 2026). Free limits change often. The catalog is checked every day.

Free plans from provider pricing pages, checked 20 November 2026
EngineFree plan (official)Cost of one searchOther limits
Tavily1,000 credits each month, no credit card1 credit (basic) or 2 credits (advanced)Extract costs 1 credit for each 5 URLs
Exa$10 in credits, reset on the first day of each month, no payment methodAbout $0.004 (lite) to $0.007 (auto)Each result above 10 costs more
Firecrawl1,000 credits each month, no credit card2 credits for each 10 results10 search requests per minute, 2 at the same time
LinkupFree credits on the pricing pageAbout $0.005 to $0.006We could not confirm the monthly amount today

With all four keys, you get about 1,000 Tavily searches, 1,400 to 2,500 Exa searches and 500 Firecrawl searches each month. These numbers are our calculation from the published prices.

Step 1: Install FreeLLMAPI and add search keys

FreeLLMAPI is open source. It runs on your computer in Docker. Your provider keys stay on your computer.

Install (macOS or Linux)

curl -fsSL https://freellmapi.co/install.sh | bash

Open the dashboard at http://localhost:3001. Add a Tavily, Exa, Firecrawl or Linkup key on the Keys page. Add a chat provider key too, for example Google AI Studio or Groq. Copy your unified key from the same page.

To see which engines can answer now, send GET /v1/search/engines. The list also gives the engine ids that you can pin.

Step 2: Search the web

Send a query. The optional fields are max_results, topic (general or news), include_answer and engine. The response has query, results, provider and engine. Each result has a title, a url and content.

curl

curl http://localhost:3001/v1/search \
  -H "Authorization: Bearer <unified-key>" \
  -H "Content-Type: application/json" \
  -d '{"query": "latest stable Python version", "max_results": 5, "topic": "general"}'

Step 3: Give the results to an LLM

The usual pattern has three steps. Search, put the results into the prompt, then ask the model to answer only from those results. Tell the model to cite the URLs.

Python: search, then answer with sources

import requests
from openai import OpenAI

BASE = "http://localhost:3001/v1"
KEY = "<unified-key>"
client = OpenAI(base_url=BASE, api_key=KEY)

question = "What is the latest stable Python version?"
hits = requests.post(
    f"{BASE}/search",
    headers={"Authorization": f"Bearer {KEY}"},
    json={"query": question, "max_results": 5},
    timeout=30,
).json()["results"]

context = "\n\n".join(
    f"[{i + 1}] {h['title']}\n{h['url']}\n{h['content'][:1500]}" for i, h in enumerate(hits)
)
answer = client.chat.completions.create(
    model="auto",
    messages=[
        {"role": "system", "content": "Answer only from the sources. Cite them as [1], [2]. "
                                      "If the sources do not answer the question, say so."},
        {"role": "user", "content": f"Sources:\n{context}\n\nQuestion: {question}"},
    ],
)
print(answer.choices[0].message.content)

Search adds time and uses credits. Search only when the question needs facts newer than the training data of the model, or when the user needs sources. Keep the results of repeated questions in a cache.

When one engine sends a 429 error, the router sends the query to the next engine. Results from all engines have the same shape, so your code does not change. Use engine only when you need one engine every time.

For an agent, give the model a web_search tool. When the model calls the tool, send the query to /v1/search and return the results as the tool result. Read Free function calling LLM API for the tool loop.

Gemini Google Search grounding

Google models can also search Google themselves. Send a tool named google_search and pin a Google model. FreeLLMAPI changes the tool into the native Gemini grounding tool.

Python: grounding with Google Search

resp = client.chat.completions.create(
    model="gemini-2.5-flash",  # pin a Google model so that the request goes to Google
    messages=[{"role": "user", "content": "Who won the F1 race this weekend?"}],
    tools=[{"type": "function", "function": {"name": "google_search", "parameters": {}}}],
)
print(resp.choices[0].message.content)
Free tier limit: Google allows grounding on the free tier for Gemini 2.5 Flash and Flash-Lite, up to 500 requests each day shared by both. Gemini 3 and newer models have no free grounding. Pin a 2.5 Flash model for free grounding.

Pitfalls and troubleshooting

  • Credits, not requests: an advanced Tavily search uses 2 credits. A Firecrawl search uses 2 credits for each 10 results. Ask for 5 results when 5 are enough.
  • Long content: search results can be long. Cut each result before you put it into the prompt, or the request can go above the token limit of a small free model.
  • Old or wrong pages: the model trusts what you give it. Tell it to say when the sources do not answer the question.
ProblemCauseFix
Error 400 query is requiredThe body has no queryAdd a query string
Error 429 with Retry-AfterAll engines are at their limitWait for the time in the header, or add another engine key
No engine is readyNo search key is addedAdd a key, then check GET /v1/search/engines
Grounding does not runThe request went to a non-Google modelPin a Gemini 2.5 Flash model
Use every one of these through one key. FreeLLMAPI is an open-source, self-hosted router that puts all these free tiers behind a single OpenAI-compatible endpoint and fails over when one is rate-limited. Browse the catalog or go live.

Frequently asked questions

Is there a free web search API for AI agents?

Yes. Tavily gives 1,000 credits each month. Exa gives $10 in credits each month. Firecrawl gives 1,000 credits each month. None of them need a credit card for the free plan.

Which free search API is best for LLMs?

Tavily can return a short answer with the results, which is useful for simple questions. Exa and Firecrawl return full page content, which is useful for research. Add all of them so that failover can use each free plan.

Is Google Search grounding free in the Gemini API?

Yes, for Gemini 2.5 Flash and Flash-Lite, up to 500 requests each day on the free tier. Gemini 3 and newer models have no free grounding.

How do I add web search to an LLM?

Search with /v1/search, put the results into the prompt, and tell the model to answer only from them and cite the URLs. For an agent, expose the search as a tool.

Sources

All guides · Browse all free models

Keep reading

All posts →
Guide

Best free LLM APIs 2026

Ten genuinely free LLM APIs worth using — ranked, with real limits.

Guide

Unlimited free AI tokens?

No free AI API is unlimited. The closest real thing: 10.1 billion free tokens a month behind one key.

Guide

How to get a free LLM API key

Get a working key in minutes — no credit card required.