SDK Setup

Wire the PulseAgent LangChain callback handler in two minutes.

Pipes every LangChain run into the PulseAgent dashboard as a parent + child trace tree. Costs, latencies, and failures show up in real time, no extra widgets to wire into your app.

1. Install

pip install pulseagent

2. Initialize (Python / LangChain)

Instantiate the handler once per process, then pass it to config={"callbacks": [handler]} on every .invoke() call. PulseAgentLangChainHandler is a BaseCallbackHandler subclass that auto-emits a parent + child trace tree as LangChain walks the chain.

import os
from langchain_openai import OpenAI
from langchain.agents import AgentExecutor, create_openai_tools_agent
from pulseagent import PulseAgentLangChainHandler

# PulseAgent auto-emits a parent + child trace tree on every callback.
handler = PulseAgentLangChainHandler(
    api_key=os.environ["PULSEAGENT_API_KEY"],
    backend_url=os.environ.get("PULSEAGENT_BACKEND_URL", "https://pulseagent-7.polsia.app"),
    agent_name="my_langchain_agent",
)

agent = AgentExecutor(agent=create_openai_tools_agent(llm=OpenAI(), tools=[], prompt=None), tools=[])

result = agent.invoke(
    {"input": "What is the capital of France?"},
    config={"callbacks": [handler]},
)
handler.flush()  # wait for trace posts to drain before process exit

3. Node / TypeScript

The TypeScript SDK has no CallbackHandler class — the equivalent is the @instrument decorator, which writes one execution event per call. Reads PULSEAGENT_API_KEY from env.

import { registerKey, instrument, flush } from "pulseagent";

// Register the API key with the backend once per process.
await registerKey({ name: "production" });

export const researcher = instrument({
  agentName: "researcher",
  model: "gpt-4o",
})(
  async function researcher(query: string) {
    const r = await openai.chat.completions.create({
      model: "gpt-4o",
      messages: [{ role: "user", content: query }],
    });
    return r;
  }
);

// Call before process exit to avoid dropping events.
process.on("SIGTERM", async () => { await flush(); process.exit(0); });

4. Verify it works

Run any LangChain AgentExecutor.invoke with the handler attached, then open the dashboard trace view to confirm the tree appears.

import os
from langchain_openai import OpenAI
from langchain.agents import AgentExecutor, create_openai_tools_agent
from langchain_core.tools import tool
from pulseagent import PulseAgentLangChainHandler

@tool
def add(a: int, b: int) -> int: return a + b

handler = PulseAgentLangChainHandler(api_key=os.environ["PULSEAGENT_API_KEY"])
agent = AgentExecutor(
    agent=create_openai_tools_agent(llm=OpenAI(), tools=[add], prompt=None),
    tools=[add],
)
agent.invoke({"input": "What is 2 + 3?"}, config={"callbacks": [handler]})
handler.flush()
# -> Click "Open Funnel tab ->" below to see the trace tree.

Open Funnel tab → In the Funnel tab click View recent agent runs to inspect the trace tree.