iflow-search-crewai

CrewAI tools for iFlow Search (心流搜索) — web search, image search, and web-page fetching, exposed as native crewai.tools.BaseTool instances.

This package is a CrewAI-native Python tool adapter. It is not a CrewAI Studio / AMP official provider tile. To appear as a first-class “iFlow Search” integration inside CrewAI’s product UI, a separate official partnership / Catalog / Marketplace route is required.

• Core SDK: iflow-search
• Sibling adapters: iflow-search-mcp, iflow-search-langchain
• API docs: https://platform.iflow.cn/docs/

Install

pip install iflow-search-crewai

Configuration

export IFLOW_API_KEY="YOUR_IFLOW_API_KEY"

Optional:

export IFLOW_BASE_URL="https://platform.iflow.cn"
export IFLOW_TIMEOUT_MS="30000"

Never commit API keys. Do not hard-code keys in source files.

Basic CrewAI usage

from crewai import Agent
from iflow_search_crewai import create_iflow_search_tools

agent = Agent(
    role="Researcher",
    goal="Research current information from the web",
    backstory="You are a careful research assistant.",
    tools=create_iflow_search_tools(),
)

Single-tool usage

from iflow_search_crewai import IFlowWebSearchTool

web_search = IFlowWebSearchTool()
result = web_search.run(query="latest AI agent frameworks", count=5)

Tools

Arguments

Each tool returns a JSON string with compact, LLM-friendly fields (titles, URLs, snippets, etc.).

DeepSeek + CrewAI smoke example

Requires crewai, this package, and API keys in the environment. Do not hard-code keys.

import os

from crewai import Agent, Crew, Task, LLM
from iflow_search_crewai import IFlowWebSearchTool

llm = LLM(
    model="deepseek/deepseek-chat",
    api_key=os.environ["DEEPSEEK_API_KEY"],
)

researcher = Agent(
    role="Web Researcher",
    goal="Answer questions using live web search",
    backstory="You use iflow_web_search when current information is needed.",
    tools=[IFlowWebSearchTool()],
    llm=llm,
    verbose=True,
)

task = Task(
    description="What is CrewAI? Give a one-sentence answer grounded in search results.",
    expected_output="A one-sentence summary with a source URL.",
    agent=researcher,
)

crew = Crew(agents=[researcher], tasks=[task], verbose=True)
result = crew.kickoff()
print(result)

If your CrewAI version prefers a different DeepSeek model string, adjust model= to match your installed CrewAI / LiteLLM configuration.

vs MCP route

Both call the same iFlow Search API via the shared iflow-search core SDK (Python) or MCP adapter (JS/Python).

Boundaries

• Not a built-in CrewAI search provider.
• Not an AMP / Studio UI tile without CrewAI official integration work.
• Does not replace iflow-search-mcp for MCP-based workflows.

Local development

cd packages/iflow-search-crewai
python -m pip install -e "../iflow-search"
python -m pip install -e ".[dev]"
python -m pytest -q
python -m ruff check .
python -m mypy src/iflow_search_crewai

Optional real-API smoke

export IFLOW_API_KEY="YOUR_IFLOW_API_KEY"
export IFLOW_CREWAI_SMOKE=1
python scripts/smoke_real_api.py