What is TinySearch? Free Web Search API for AI Agents
A pricing comparison sounds simple until your AI application has to figure out where the right information actually lives.
Before it can compare plans, it needs to find the official pricing pages. Before it can research a company, it needs to locate the right documentation, announcements, filings, or other relevant sources. The reasoning can only begin once the application has found something worth reading.
That discovery step is where TinySearch fits!
TinySearch is TinyFish’s web search API for AI agents and applications. It takes a query and returns structured search results that help an application identify the most relevant pages and decide what to read next.
This guide explains how TinySearch works, what it returns, and how it fits into the wider TinyFish platform.
What is TinySearch?
TinySearch is the search layer of the TinyFish platform. It is designed to help AI agents and applications discover relevant information on the web.
An application sends TinySearch a query and receives structured search results including the page title, URL, snippet, result position, and other available source information. The results can then be evaluated by the application, passed into a language model, or used as the starting point for another web operation.
The important distinction is that TinySearch finds the source.
It does not need to retrieve and process the entire page to help an application decide where useful information is likely to be. If the application later needs the content of one of those pages, it can pass the URL to TinyFetch.
This makes TinySearch useful on its own, but it can also act as the first step in a larger workflow. Search is a way to find pages before retrieving their content with TinyFetch.
What problem does TinySearch solve?
Large language models can reason over the information they receive, but that does not mean every application automatically knows where to find the right information on the web.
Imagine an agent gets this task:
“Find the current pricing for these five project management platforms.”
The application first needs to determine which pages are relevant.
Searching manually is easy for a person. You enter a query, scan the results, select the source that looks useful, and continue.
An AI application needs a programmatic version of that process.
TinySearch gives the application a way to submit a search query and receive results that software can work with directly. Instead of returning a page designed for a person to browse, the API returns structured search information that the surrounding application can evaluate.
For example, an agent researching a company might use TinySearch to find:
- the company’s official pricing page;
- its latest product announcement;
- relevant documentation;
- an investor relations page;
- or sources discussing a specific topic.
The application remains responsible for deciding which result is useful and what to do with it next. TinySearch handles the discovery step.
How does TinySearch work?
At a high level, a TinySearch workflow has three parts: query, results, next action.
1. The application submits a search query
The process begins with the information the application needs.
For example:
Acme project management pricing
Or:
Acme API authentication documentation
TinySearch accepts the query and searches for relevant web results. Developers can narrow results by location or language, or use recency and date filters for time-sensitive searches, as described in the Search API reference.
The query does not need to contain a URL. The purpose of search is to help the application discover where the relevant information lives.
2. TinySearch returns ranked results
TinySearch returns structured results rather than an answer written in natural language.
A result can contain fields such as:
{
"position": 1,
"site_name": "example.com",
"title": "Pricing",
"snippet": "Plans for teams of all sizes...",
"url": "https://example.com/pricing"
}The application can use these fields to determine which sources are worth opening or processing further.
The Search API reference documents these result fields and notes that publication dates appear only for some results.
3. The application decides what happens next
Finding a relevant page may be enough for some workflows.
For others, search is only the beginning.
If the agent needs a result’s page content, the application can send the URL to TinyFetch. For multi-step website actions, use TinyAgent; use TinyBrowser when the application will control a managed browser session itself.
That separation is intentional: search, retrieval, browser infrastructure, and website operation are different jobs.
What does TinySearch return?
TinySearch is designed to give applications enough information to identify and evaluate web sources without forcing the model to process an entire search-results page.
| TinySearch result | What it tells the application |
|---|---|
| Title | What the page appears to be about |
| URL | Where the original source is located |
| Snippet | A relevant excerpt that helps assess the result |
| Position | Where the result appears in the ranked results |
| Site information | Additional context about the source |
The snippet is useful for deciding whether a page is relevant, but it is not a substitute for the full page.
Suppose an agent searches for a product’s pricing. A result snippet might mention the starting price, but omit billing frequency, usage limits, enterprise terms, or other details available on the page.
If those details matter, that is where retrieval comes in.
TinySearch helps find the page. TinyFetch helps read the page.
Search and Fetch give developers a simple starting point: find a relevant URL, then retrieve its content. More complex tasks can use TinyFish’s browser infrastructure or web agent when reading alone is not enough.
What would your agent search for?
Try a real query and see the ranked sources TinySearch returns.
→ Run it in the Playground
What can you use TinySearch for?
TinySearch can fit into any application that needs to discover web sources from a query. Here are a few common workflow shapes.
Research assistants
Research rarely begins with a perfect list of URLs.
An AI research assistant might receive a question, search for several relevant sources, select the most useful pages, retrieve their contents, and then reason across the evidence.
TinySearch handles the discovery part of that process.
For example, an application researching a new software category could search for company documentation, product pages, recent announcements, and third-party sources before deciding which pages need deeper analysis.
The application performs the final synthesis. TinySearch supplies the web-search layer that helps it find the inputs.
Finding information across multiple websites
Some tasks require the same information from several independent sources.
An application comparing laptop prices, for example, might first search for relevant retailer or manufacturer pages. A market-intelligence workflow might search for company announcements across dozens of organizations.
TinySearch can locate the relevant sources before the application moves into retrieval.
This is especially useful when the application knows what it needs but does not already know where the information is located.
Recurring web research
Search can also be part of a repeating workflow.
A system might regularly look for newly published sources around a market, company, product category, policy topic, or other subject. TinySearch can perform the source-discovery step each time the workflow runs.
The scheduling, comparison, notification, or final analysis belongs to the surrounding application or the appropriate monitoring capability. TinySearch itself remains focused on finding relevant web results.
What makes TinySearch useful for AI applications?
Traditional search is designed primarily around a person looking at a search-results page.
AI applications need something slightly different: search results that software can consume and pass into the next step of a workflow.
Structured results instead of a search page
TinySearch returns structured data, including titles, URLs, and snippets, so applications do not need to turn a human-facing search page into usable fields first.
That makes it easier to connect search directly to an LLM, retrieval pipeline, data workflow, or another TinyFish capability. TinyFish’s current product page describes TinySearch as returning structured JSON for agents.
Search controls for more specific queries
Not every search should operate across the entire web without constraints.
TinySearch supports location and language parameters for region-specific results. It also supports recency and date filters when an application needs sources from a particular period.
For example, a developer could constrain a search to more recent sources when building a workflow around news or fast-changing information.
A clear path from discovery to retrieval
Search often becomes more useful when the application can immediately do something with the URL it discovers.
TinyFish separates these jobs instead of treating every web task as the same operation.
An application can search first, retrieve only the pages that matter, and move into browser or agent-based operations only when the task actually requires them.
That matters because most workflows do not need the full web stack at every step.
TinySearch vs. TinyFetch vs. TinyAgent vs. TinyBrowser
The easiest way to understand TinySearch is to compare it with the other TinyFish primitives by the job each one performs.
| TinyFish product | Use it when you need to… |
|---|---|
| TinySearch | Find relevant web sources from a query |
| TinyFetch | Retrieve content from a URL you already know |
| TinyAgent | Give TinyFish a goal and have it complete a multi-step website task |
| TinyBrowser | Control a managed cloud browser from your own automation code |
These products can be used independently.
If your application already knows the URL, start with TinyFetch to retrieve its content.
If the application only needs relevant sources, TinySearch may be enough.
If the job evolves into navigating pages, working through filters or pagination, filling forms, or completing an authorized website workflow, it moves beyond search and retrieval into TinyAgent.
TinyBrowser serves a different developer need: managed browser infrastructure for teams that want to control the browser themselves rather than hand TinyFish a goal.
TinyFish describes this progression as moving from find, read, and extract into website operation as the job becomes more complex.
Already know where the information lives?
Skip search and use TinyFetch to retrieve the page directly.
→ Try TinyFetch
Is TinySearch free?
Yes. TinySearch is free to use.
TinySearch and TinyFetch are free within their published limits and do not draw from the prepaid Wallet. TinyAgent and TinyBrowser draw from Wallet funds when used.
That makes TinySearch an easy starting point for developers who want to add web search to an AI agent or application before building out more complex retrieval or web automation workflows.
How to get started with TinySearch
You need a TinyFish API key before making your first request.
With a TinyFish API key configured, the Search API reference shows this Python SDK pattern:
from tinyfish import TinyFish
client = TinyFish()
response = client.search.query(
query="web automation tools"
)
for result in response.results:
print(result.title, "→", result.url)The API returns structured results that your application can inspect, rank further, send to an LLM, or pass into another step of the workflow.
The TinyFish documentation also includes TypeScript and cURL examples, along with authentication, request parameters, filtering options, and the full Search API reference.
Start with search, then go as far as the task requires
Not every AI application needs to operate a browser or automate a website.
Sometimes the first problem is much simpler: find the right source.
TinySearch gives applications a programmatic search layer for doing exactly that. It turns a search query into structured web results that an application can evaluate and use in the next step.
For a research assistant, that might mean finding the source and passing it to TinyFetch. For a larger workflow, search may eventually lead into extraction, browser access, or multi-step website operations.
Start with TinySearch when the application needs to find a page; use TinyFetch when it already has the URL and needs the page content.
Search is free. Test it with your own workflow.
Give TinySearch a query and see how it could fit into your agent or application.
→ Try TinySearch in the Playground
Related Reading:
AI disclosure
Content on this website may be created or refined with the assistance of AI tools and is subject to human editorial review.
