Skip to main content
Zespan is an agent observability platform that helps you understand exactly what your AI agents are doing in production. Instrument your agents with two lines of code, then explore real-time traces, costs, tool call histories, and AI-powered insights in the Zespan dashboard.

Quickstart

Instrument your first agent and see a trace in the dashboard in under 5 minutes.

Core concepts

Understand traces, spans, span kinds, projects, and the agent observability data model.

TypeScript SDK

Wrap OpenAI, Anthropic, Google, and other clients. Full agent tracing with withAgent.

Python SDK

Patch your Python LLM clients and use with_agent for multi-step agent tracing.

LangChain

Full chain, agent executor, and retriever tracing with LumiqtraceCallbackHandler.

Google ADK

Multi-agent tracing with instrumentADK, wrapADKRunner, or wrapADKAgent.

​
Get started in 3 steps

1

Create a project and get your API key

Sign up at app.zespan.com, create an organization, and create your first project. Your project API key (zsp_...) is shown once on creation — copy it now.
2

Install the SDK and instrument your agent

Install @zespan/sdk (Node.js) or zespan (Python), initialize with your API key, and wrap your existing clients or agent frameworks. No changes to your agent logic required.
import { zespan, withAgent } from "@zespan/sdk";
import OpenAI from "openai";

zespan.init({ apiKey: process.env.ZESPAN_API_KEY!, environment: "production" });
const openai = zespan.wrapOpenAI(new OpenAI());
3

View your agent trace in the dashboard

Run your agent. Within seconds, the full trace appears in Zespan with a flame graph showing every agent turn, tool call, and LLM interaction — with cost and latency for each step.
Zespan adds zero latency to your agent runs. All telemetry is batched and flushed asynchronously — your agents are never blocked waiting for observability data to be sent.