langchain-autom

LangChain integration for Autom, a SERP API that returns live search engine results as JSON.

It adds two components:

• AutomSearch, a tool that agents call to search the web;
• AutomSearchRetriever, a retriever that returns the results as LangChain documents.

Both work with Google (the default), Bing, Brave, DuckDuckGo and Yahoo.

Installation

pip install -U langchain-autom

Get an API key at app.autom.dev and set it in the environment:

export AUTOM_API_KEY="your-api-key"

You can also pass api_key=... to either component.

Search tool

from langchain_autom import AutomSearch

tool = AutomSearch(max_results=5)
tool.invoke({"query": "latest LangChain release"})

{
    "query": "latest LangChain release",
    "results": [{"title": "...", "url": "https://...", "content": "...", "position": 1}, ...],
}

The model sets query, and optionally time_range (day, week, month or year) to keep only recent pages.

You set the other options when you create the tool:

Option	Default	Description
engine	google	google, bing, brave, duckduckgo or yahoo
max_results	5	Results returned to the model, 1 to 10
country		Two-letter country code, such as us (Google only)
language		Language code of the results, such as en (Google only)
location		City or region to search from, such as Paris, France (Google only)
timeout	60	Seconds to wait for Autom

API errors are returned to the model as text instead of raising, and the key never appears in them. Autom reports a rate limit as HTTP 401 with "Rate limit exceeded"; the tool says "rate limit exceeded" in that case.

With an agent

from langchain.agents import create_agent
from langchain_autom import AutomSearch

agent = create_agent("openai:gpt-4.1-mini", tools=[AutomSearch()])
agent.invoke({"messages": [{"role": "user", "content": "What changed in Python 3.13?"}]})

Retriever

from langchain_autom import AutomSearchRetriever

retriever = AutomSearchRetriever(k=5, engine="brave")
docs = retriever.invoke("retrieval augmented generation")

Each document's page_content is the result snippet. Its metadata holds title, url and position, plus date and source when Autom returns them. The retriever takes the same options as the tool, with k in place of max_results and a fixed time_range.

Both components have async versions (ainvoke).

Development

pip install -e ".[test]"
pytest tests/unit_tests                      # offline
AUTOM_API_KEY=... pytest tests/integration_tests  # live

License

MIT