Ferret + LangChain Integration

Ferret ships LangChain-compatible tools that wrap the REST API. They turn Ferret into a drop-in search/research backend for any LangChain agent.

Tools

Tool	Description	Endpoint
`FerretSearchTool`	Multi-engine web search	`GET /search`
`FerretResearchTool`	Deep research with aggregation + summary	`POST /research`
`FerretExtractTool`	Extract raw content from a URL	`GET /extract`
`FerretCrawlTool`	Crawl a URL for raw content	`GET /crawl/v1`

All tools return Tavily-compatible payloads, so existing LangChain agents built against Tavily can be migrated by swapping the tool import.

Quick start

from langchain_ferret import FerretSearchTool, FerretResearchTool

# Search
search = FerretSearchTool()
results = search.run("latest AI news")
print(results)

# Research
research = FerretResearchTool()
answer = research.run("Impact of quantum computing on cryptography")
print(answer)

Usage with a LangChain agent

from langchain.agents import initialize_agent, AgentType
from langchain_ferret import FerretSearchTool, FerretExtractTool
from langchain_openai import ChatOpenAI

tools = [FerretSearchTool(), FerretExtractTool()]
llm = ChatOpenAI(model="gpt-4")
agent = initialize_agent(tools, llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION)
agent.run("Search for latest AI news and extract the content from the top result")

Extract and crawl

from langchain_ferret import FerretExtractTool, FerretCrawlTool

extract = FerretExtractTool()
content = extract.run("https://example.com/article")

crawl = FerretCrawlTool()
site = crawl.run("https://example.com")

Business endpoints as tools

The 17 business endpoints can be wrapped in custom tools. The /llm variants return markdown prompts ready for direct LLM injection:

import requests
from langchain.tools import BaseTool

class FerretSEOAuditTool(BaseTool):
    name: str = "ferret_seo_audit"
    description: str = "Run a technical SEO audit of a site. Input: domain."

    def _run(self, site: str) -> str:
        # The /llm variant returns an LLM-ready markdown prompt (~800 tokens)
        r = requests.get(
            "http://localhost:9093/seo/audit/llm",
            params={"q": site},
            timeout=60,
        )
        return r.text

    async def _arun(self, site: str) -> str:
        return self._run(site)

Requirements

pip install langchain requests
# Optional, for OpenAI-based agents:
pip install langchain-openai

Make sure Ferret is running on http://localhost:9093:

systemctl --user start ferret.service    # REST API on :9093