LangChain / LangGraph

Add Trappr security to any LangChain or LangGraph agent in two lines of Python or TypeScript. Route all LLM calls through the Trappr Gateway — no changes to your chains, agents, or tools.

How it works

LangChain's ChatOpenAI and ChatAnthropic classes (and their equivalents) accept a custom base_url. Point that at the Trappr Gateway and swap your provider key for your Trappr Gateway key — the rest of your agent code is untouched.

LangGraph Agent

invoke / stream

→

ChatOpenAI

base_url overridden

→

Trappr Gateway

canary · DLP · log

→

OpenAI / Anthropic

LLM provider

Python — LangChain / LangGraph

ChatOpenAI

Python

from langchain_openai import ChatOpenAI

# Before
llm = ChatOpenAI(model="gpt-4o", openai_api_key="sk-...")

# After — only openai_api_base and openai_api_key change
llm = ChatOpenAI(
    model="gpt-4o",
    openai_api_key="<your-trappr-gateway-key>",
    openai_api_base="https://gateway.trappr.net/v1",
)

ChatAnthropic

Python

from langchain_anthropic import ChatAnthropic

# Use the OpenAI-compatible shim for Anthropic via Trappr
from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    model="claude-sonnet-4-6",
    openai_api_key="<your-trappr-gateway-key>",
    openai_api_base="https://gateway.trappr.net/v1",
)

Full LangGraph agent example

Python

import os
from langchain_openai import ChatOpenAI
from langgraph.graph import StateGraph, END
from typing import TypedDict, Annotated
import operator

TRAPPR_KEY = os.environ["TRAPPR_GATEWAY_KEY"]

llm = ChatOpenAI(
    model="gpt-4o",
    openai_api_key=TRAPPR_KEY,
    openai_api_base="https://gateway.trappr.net/v1",
)

class AgentState(TypedDict):
    messages: Annotated[list, operator.add]

def call_model(state: AgentState):
    response = llm.invoke(state["messages"])
    return {"messages": [response]}

graph = StateGraph(AgentState)
graph.add_node("agent", call_model)
graph.set_entry_point("agent")
graph.add_edge("agent", END)

app = graph.compile()
result = app.invoke({"messages": [("user", "Summarize this report.")]})

TypeScript / JavaScript

TypeScript

import { ChatOpenAI } from "@langchain/openai";

// Before
const llm = new ChatOpenAI({ model: "gpt-4o", openAIApiKey: "sk-..." });

// After
const llm = new ChatOpenAI({
  model: "gpt-4o",
  openAIApiKey: process.env.TRAPPR_GATEWAY_KEY,
  configuration: {
    baseURL: "https://gateway.trappr.net/v1",
  },
});

Execution tracking with LangChain

For multi-step agents, wrap the execution with a Trappr execution token so all LLM calls from one agent invocation are grouped together in the dashboard.

Python

import httpx

def start_execution(agent_key: str, agent_id: str) -> str:
    resp = httpx.post(
        "https://api.trappr.net/v1/agents/executions",
        headers={"Authorization": f"Bearer {agent_key}"},
        json={"agentId": agent_id},
    )
    return resp.json()["token"]

def end_execution(agent_key: str, token: str, status: str):
    httpx.patch(
        f"https://api.trappr.net/v1/agents/executions/{token}",
        headers={"Authorization": f"Bearer {agent_key}"},
        json={"status": status},
    )

# Usage
token = start_execution(AGENT_KEY, "cht_agent_XXXX")

llm = ChatOpenAI(
    model="gpt-4o",
    openai_api_key=AGENT_KEY,
    openai_api_base="https://gateway.trappr.net/v1",
    default_headers={"X-Trappr-Execution": token},
)

try:
    result = app.invoke({"messages": [("user", "Process this data.")]})
    end_execution(AGENT_KEY, token, "completed")
except Exception as e:
    end_execution(AGENT_KEY, token, "failed")
    raise

Environment variable setup

Shell

# .env
TRAPPR_GATEWAY_KEY=cht_gw_XXXX
TRAPPR_AGENT_KEY=cht_ak_XXXX
TRAPPR_AGENT_ID=cht_agent_XXXX

# Remove or comment out your direct provider keys
# OPENAI_API_KEY=sk-...   <- no longer needed

If LangChain reads OPENAI_API_KEY from the environment automatically, set it to your Trappr gateway key and set OPENAI_API_BASE tohttps://gateway.trappr.net/v1. LangChain will pick both up without any code change.

LangChain callbacks

Trappr is fully compatible with LangChain callbacks. You can continue using LangSmith, Lunary, or any other callback handler alongside Trappr monitoring — they operate independently.

Python

from langchain.callbacks import LangChainTracer

llm = ChatOpenAI(
    model="gpt-4o",
    openai_api_key=TRAPPR_KEY,
    openai_api_base="https://gateway.trappr.net/v1",
    callbacks=[LangChainTracer()],  # works alongside Trappr
)

← PreviousMake (Integromat)Next →Zapier