langchain-asterwise

LangChain tools for the Asterwise astrology API. Five StructuredTools over the official Python SDK:

Outputs are slimmed by default to the fields a model needs to reason with. Pass slim=False for the full API payload. Birthplaces are plain text; the API geocodes them.

Install

pip install langchain-asterwise

Get a free key at asterwise.com/dashboard: 500 calls a month, no card.

Use

import os
from asterwise_langchain import AsterwiseToolkit

tools = AsterwiseToolkit(api_key=os.environ["ASTERWISE_API_KEY"]).get_tools()

With any LangChain chat model that supports tool calling:

from langchain.chat_models import init_chat_model

llm = init_chat_model("claude-sonnet-5").bind_tools(tools)
msg = llm.invoke("Cast a Vedic chart for 12 Nov 1985, 06:45, Mumbai and tell me the Moon nakshatra.")
print(msg.tool_calls)

Or as a ReAct agent with LangGraph:

from langgraph.prebuilt import create_react_agent

agent = create_react_agent(init_chat_model("claude-sonnet-5"), tools)
result = agent.invoke({"messages": [("user", "Are these two compatible? Arjun 1990-05-14 07:20 Pune; Meera 1992-11-03 22:45 Jaipur.")]})
print(result["messages"][-1].content)

The toolkit's methods also work without LangChain:

kit = AsterwiseToolkit(api_key="aw_...")
kit.matchmaking("1990-05-14", "Pune, India", "1992-11-03", "Jaipur, India")

Why the matchmaking tool reports vetoes separately

Rajju and Vedha between the two Moon nakshatras are stops in the classical method, not point deductions. A 30 out of 36 with Rajju present is not a good match, so the tool returns classical_vetoes alongside total_score and the tool description tells the model so. Background: Rajju and Vedha as hard vetoes.

Accuracy

Every Asterwise position is computed with the Swiss Ephemeris and checked against NASA JPL Horizons: 80 positions from 1950 to 2050, median difference 0.046 arcseconds, raw data and script published.

Test

pip install -e ".[test]" && pytest            # offline, mocks the SDK
ASTERWISE_API_KEY=aw_... python scripts/smoke.py   # one live call per tool

MIT licensed.