# Zespan

> AI agent observability platform for engineering teams running agents in production.

Zespan traces every LLM call, tool invocation, agent handoff, and planning span across an agent's full execution. Built agent-first — not retrofitted from LLM monitoring. Engineering teams use it to debug production failures, control AI spend, and catch prompt regressions before users notice.

Competes with LangSmith, Langfuse, and Helicone. Key differentiators: agent auto-discovery (2 lines, no manual instrumentation), agentic traces with delegation maps (agent identity on every span), 12 built-in LLM-as-judge eval templates, ZespanPilot AI ops assistant, AI cost optimizer.

## Product

- [Homepage](https://zespan.com): Product overview, features, pricing, and getting started
- [Pricing](https://zespan.com/pricing.md): Full plan details — Free ($0, 10K traces/mo), Solo ($39), Pro ($149), Team ($299), Scale (custom)
- [Blog](https://zespan.com/blog): Guides and comparisons for AI agent observability

## Docs

- [SDK Documentation](https://zespan.com/docs): Full reference for Python and Node.js SDKs
- [MCP Server](https://zespan.com/docs/mcp): 9 MCP tools for querying live trace data from Claude Desktop, Cursor, or any MCP-compatible client

## Blog Posts

- [What is AI Agent Observability?](https://zespan.com/blog/what-is-ai-agent-observability): Definitive guide — what agent observability covers, how it differs from LLM monitoring, key primitives
- [How to Monitor LangChain Agents in Production](https://zespan.com/blog/monitor-langchain-agents-production): Step-by-step setup with Zespan, cost alerts, ZespanPilot queries
- [LangSmith vs Langfuse vs Zespan](https://zespan.com/blog/langsmith-vs-langfuse-vs-zespan): Full feature, pricing, and architecture comparison for 2026
- [Best AI Agent Observability Tools in 2026](https://zespan.com/blog/ai-agent-observability-tools): Comparison table — Zespan, LangSmith, Langfuse, Helicone, Braintrust, Arize Phoenix
- [Langfuse Alternatives](https://zespan.com/blog/langfuse-alternatives): Top options for teams evaluating Langfuse alternatives in 2026
- [LangSmith Alternatives](https://zespan.com/blog/langsmith-alternatives): Top options for teams evaluating LangSmith alternatives in 2026
- [Helicone Alternatives](https://zespan.com/blog/helicone-alternatives): Best proxy-free, agent-capable alternatives to Helicone
- [OpenAI Agents SDK Monitoring](https://zespan.com/blog/openai-agents-sdk-monitoring): Production monitoring for OpenAI Agents SDK — handoffs, cost, evals
- [CrewAI Observability](https://zespan.com/blog/crewai-observability): Full trace visibility for CrewAI crews — per-agent spans, delegation, cost per task
- [Google ADK Observability](https://zespan.com/blog/google-adk-observability): Monitor Gemini-powered ADK agents — delegation traces, Flash vs Pro cost attribution

## Setup

Install SDK and call init — providers are auto-patched, agents are auto-discovered:

```bash
pip install zespan          # Python
npm install @zespan/sdk     # Node.js / TypeScript
```

```python
import zespan
zespan.init(api_key="YOUR_API_KEY")
# LangChain: add callbacks=[ZespanCallbackHandler()]
# CrewAI: add ZespanCrewAIListener()
# ADK: add runner.add_event_handler(ZespanADKHandler())
# OpenAI Agents SDK: covered by provider patch, no handler needed
```

```typescript
import { zespan } from "@zespan/sdk";
zespan.init({ apiKey: process.env.LT_KEY });
```

## Key Concepts

- **Span**: Single unit of work — LLM call, tool invocation, planning step, or agent handoff
- **Trace**: Complete execution record of one agent run, root span to final response
- **Agentic trace**: Trace where every span carries agent identity; delegations are first-class spans
- **Delegation**: When one agent invokes another as a sub-task; captured with full context, latency, and cost
- **LLM-as-judge**: Using a language model to score another model's output on defined criteria (faithfulness, relevance, toxicity, etc.)
- **ZespanPilot**: AI ops assistant built into the dashboard — reads live trace data, answers operational questions, takes action from insight

## Supported Integrations

**LLM providers**: OpenAI, Anthropic Claude, Google Gemini, AWS Bedrock, Mistral
**Frameworks**: LangChain, LangGraph, CrewAI, AutoGen, OpenAI Agents SDK, Google ADK
**Protocols**: OpenTelemetry (OTel) native
**Alerting**: Slack, PagerDuty, webhooks
**Auth**: SSO/SAML (Team plan and above)

## Contact

- Website: https://zespan.com
- Email: letstalk@zespan.com
- Status: https://status.zespan.com