langchain-alion

LangChain tools for Alion: a live graph of companies worldwide (what each one does, where it operates, its tech stack, funding rounds and investors, how fast it hires), startup investors, verified job postings, and hiring and pay statistics.

Give an agent prospect lists, company enrichment by domain, lookalikes, investor research and market pay data. Most tools are free without an account.

pip install langchain-alion

Quick start

from langchain.agents import create_agent
from langchain_alion import AlionToolkit

agent = create_agent(
    "anthropic:claude-sonnet-5-5",
    tools=AlionToolkit().get_tools(),
)
result = agent.invoke({"messages": [{"role": "user", "content": "Find seed-stage fintech startups in Germany that are hiring, and their tech stacks."}]})

Any chat model that supports tool calling works. The tools also work in LangGraph and on their own:

from langchain_alion import AlionSearchCompanies

tool = AlionSearchCompanies()
print(tool.invoke({"industry": "Fintech", "country": "DE", "startup": True, "funding_stage": "seed", "limit": 5}))

Free without a key, everything with a key

Without a key, AlionToolkit().get_tools() returns the tools that are free on Alion's open address: search companies, similar companies, resolve a company, check a counterparty, search job postings and get a job posting. The allowance is 300 calls a day per network and up to 20 rows per list.

With an API key you get every tool, including full company and investor profiles and pay statistics:

export ALION_API_KEY=...

or AlionToolkit(api_key="..."). Calls with a key are paid from your account's credits (prices).

Tools

Pick a subset with AlionToolkit(include=["search_companies", "get_company"]). Arguments and descriptions come from the Alion server's own tool schemas.

Each tool returns the answer as compact JSON text for the model, and the parsed answer as the message artifact (response_format="content_and_artifact"). When the server refuses a call (a paid tool without a key, a rate limit), the reason goes back to the model as text, so the agent can adjust instead of failing.

Options

AlionToolkit(
    api_key="...",            # default: ALION_API_KEY
    app_name="my-agent/1.0",  # added to the User-Agent
    timeout=120.0,            # seconds per call
    include=None,             # Alion tool names to include
)

Links

• Alion docs: https://alion.io/developers
• Python client this package builds on: alion
• MCP server (for clients that speak MCP directly): https://alion.io/mcp/open, or npx -y @alion-io/mcp
• Source and issues: https://github.com/alion-io/alion-sdk
• MIT license