Migrate a LangChain agent from Tavily to Decodo's Web Data API

This example replaces the Tavily search tool in a LangChain agent with 2 tools, web_search and fetch_page, backed by Decodo's Web Data API. Both tools call one retrieval interface, so switching the backend is a config change, and Tavily stays available as a fallback.

It's the complete, tested code from the article Tavily alternatives: 7 agentic search APIs compared.

What you get

• Search and page fetch behind one API key. The web_search tool returns live Google results, and the fetch_page tool returns the page itself as markdown.
• Pages that search alone can't reach. Fetches start on the lowest-cost settings and add the premium pool, then JavaScript rendering, only when a page needs it.
• The Tavily options your agent uses. Domain and date filters carry over, so answers don't change without warning.
• A safe switch. Run both backends side by side on your own URLs before you commit.

Quickstart

You need Python 3.10 or later and a Web Data API key. Get your key on the free plan, then copy .env.example to .env and add the key.

With uv, one command installs the dependencies and runs the demo:

uv run --env-file .env agent_tools.py

With pip:

pip install -r requirements.txt
set -a; source .env; set +a
python agent_tools.py

The demo runs one web search and fetches one page as markdown.

Use the tools in your agent

Swap the Tavily tool for the 2 new tools wherever you build your agent's tool list.

# Before
from langchain_tavily import TavilySearch
tools = [TavilySearch(max_results=5)]

# After
from agent_tools import web_search, fetch_page
tools = [web_search, fetch_page]

The model can pass the same filters it used with Tavily.

web_search.invoke({"query": "langchain release", "domains": ["github.com"], "time_range": "month"})

Check your setup

python smoke_test.py

It runs a search and a page fetch with your key, then prints "All checks passed".

Files

File	What it does
retrieval.py	The retrieval interface. The DecodoRetriever class searches with the Google Search target and fetches pages with /scrape. The TavilyRetriever class keeps your existing Tavily setup behind the same interface.
agent_tools.py	The LangChain tools web_search and fetch_page, plus a short demo.
shadow_eval.py	Sends the same URLs to both backends and labels each result as complete, partial, blocked, or error.
smoke_test.py	Confirms that your key and setup work.
urls.txt	Sample URLs for the shadow evaluation. Replace them with URLs from your own agent logs.

Tavily options and their Web Data API equivalents

Tavily option	Web Data API equivalent
include_domains	site: operators in the query, with google_nfpr set to true so Google keeps the exact query
time_range	google_tbs, for example qdr:w for the past week
include_raw_content	A fetch_page call for the result URL
country	geo, which runs the search from that country instead of boosting results
include_answer	The agent's own model, which writes the answer from full pages instead of snippets

Run a shadow evaluation

Put the URLs your agent fetched in production into urls.txt, one per line, then run:

python shadow_eval.py urls.txt

With TAVILY_API_KEY set, each URL is checked on both backends. Compare completeness first, because a CAPTCHA page passed to the model as content can change the agent's answer.

Keep Tavily as a fallback during the switch

Leave your Tavily setup behind the same interface for 1 release cycle, so switching backends stays a one-line config change. To route calls to it, install the optional extra and set 2 variables.

pip install tavily-python
export TAVILY_API_KEY=your-tavily-key
export RETRIEVER=tavily

Learn more

• Web Data API documentation
• Pricing
• Decodo LangChain integration