langchain-relm

LangChain tools for Relm - the CRM for LLMs and AI agents.

Give a LangChain / LangGraph agent a real CRM: create and find contacts, companies and deals, log activities, and move deals through a pipeline - in a few typed tools. Relm's errors are RFC-9457 problem+json with valid_options and a suggestion, and this adapter passes them straight back to the model, so your agent self-corrects in one turn instead of crashing on a bad field or stage.

Install

pip install langchain-relm                                  # once it's on PyPI
# or install the latest straight from source:
pip install "git+https://github.com/98000m/langchain-relm.git"

Quickstart

Mint a free key (and a free test-mode key) at app.relmcrm.com, then:

from langchain_relm import get_relm_tools
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent

tools = get_relm_tools(api_key="relm_live_...")   # or set RELM_API_KEY
agent = create_react_agent(ChatOpenAI(model="gpt-4o"), tools)

agent.invoke({"messages": [(
    "user",
    "Add Maria Diaz (maria@acmelogistics.com) as a lead, create Acme Logistics "
    "as her company, and open a $18,500 deal for the annual plan.",
)]})

The agent calls relm_describe_schema to learn what exists, then creates the company, the contact and the deal - and if it sends a stage or type that doesn't exist, Relm hands back the valid options and it retries.

The tools

Direct client

Skip the tools and call the API yourself:

from langchain_relm import RelmClient

relm = RelmClient(api_key="relm_live_...")
relm.get("/schema")
relm.post("/contacts", {"email": "ada@example.com", "type": "lead"})

Notes

• Test mode is free. A relm_test_ key writes to an isolated, unmetered dataset - rehearse there, then swap one string.
• Native MCP + A2A too. Prefer MCP? Relm ships a native MCP server at https://api.relmcrm.com/mcp and an A2A endpoint - see the agent hub.
• Docs: relmcrm.com/docs · OpenAPI: relmcrm.com/openapi.json

License

MIT