Engineering & AI
TinyFish: choosing a tool when an AI agent needs web data
TinyFish gives AI agents tools to search, read, and interact with websites so they can collect data even when it appears only after a few steps.
Summary
- TinyFish gives an AI agent ways to find sources, read pages and operate a browser to retrieve web data.
- Try TinyFetch when the data is available at a known URL. For workflows with filters or pagination, choose TinyBrowser or TinyAgent based on who should decide each step.
- Before importing results, reopen the source URL and verify the fields that drive your next decision.

You want to import prices from a partner catalog, but the site only shows the right price after you choose a product line and region. An AI agent that reads the first page may pick up the default price for the wrong region. TinyFish offers four tools for different parts of that job. Choose based on what the page actually shows and what it takes to reach the data.
Choose by the work your agent needs to do
- You need to find the source. TinySearch returns URLs, titles and snippets. Use it to find the partner's official catalog, then open the catalog for the price. TinySearch documentation
- You already have the URL that shows the data. TinyFetch returns page content as Markdown, HTML or JSON. If the catalog gives you a shareable URL with filters applied, start by reading that URL. TinyFetch documentation
- Your team already has a fixed script. TinyBrowser provides a remote browser session your code can control with Playwright or a compatible tool. Your script decides what to click and what to check. TinyBrowser documentation
- The next step depends on what the page shows. Give TinyAgent a starting URL and a goal. It can choose actions such as finding a filter and opening individual rows to read prices. How TinyAgent works
Try one product group first
Apply both filters in the partner catalog, then copy the URL into a new tab. If the filters survive and TinyFetch returns the right prices, the team can use that URL. If the data requires interaction, try reading three products from a single catalog the team is authorized to use. With TinyAgent, output_schema defines the fields and data types you expect. Declaring output for TinyAgent
Preparing the AI agent
- Provide a starting URL and scope. Specify the catalog, region, product line and a limit of three results. Keep the trial read-only.
- Name the fields to collect. Each row should include a product code, price, currency and source URL. Leave any field missing from the page empty and record why.
- Prepare reference rows. Manually collect one regional price, one item with no price and one item on a later page.
Check the result before using it
A COMPLETED status means the run has finished. Before importing the data, reopen the source URL and compare product code, region, price and currency. Keep rows with missing prices separate for follow-up. Review limits and costs before increasing run volume. The trial above is illustrative. TinyFish pricing
Terms in this article (2)
- TinyFish
- A platform for finding, reading and interacting with web pages.
- output_schema
- A declaration of the fields and types TinyAgent should return. Downstream code still needs to validate the values.
Frequently asked questions
When is TinyFetch enough for a filtered catalog?
Use it when reopening the URL restores the filters and the returned page content includes the fields you need.
When should I choose TinyBrowser over TinyAgent?
Choose TinyBrowser when the team wants a fixed script that controls each browser step and verifies the result.
Sources
- TinyFish Developer Documentation, TinyFish
- TinySearch, TinyFish
- TinyFetch, TinyFish
- TinyBrowser, TinyFish
- What is TinyAgent?, TinyFish
- TinyFish Pricing, TinyFish