# Langfuse Python SDK — Trace LLM Apps

> Langfuse Python SDK adds tracing and observability to any LLM app via decorators or low-level calls, so you can track latency, cost, and prompts.

## Install

Copy the content below into your project:

## Quick Use

```bash
pip install langfuse
```

Then follow the SDK docs (v4 rewrite released March 2026) for your framework integration.

## Intro

Langfuse Python SDK adds tracing and observability to any LLM app via decorators or low-level calls, so you can track latency, cost, and prompts.

- **Best for:** Python LLM apps that need reliable tracing across prompts, tools, and providers
- **Works with:** Python; decorators or low-level events; works with any LLM/provider (per README)
- **Setup time:** 5–20 minutes

## Practical Notes

- Per README: SDK v4 rewrite shipped in March 2026 (check the v4 migration guide before upgrading).
- Start with one endpoint/function, then expand tracing to tool calls and background jobs.
- Log only what you can keep: scrub secrets and PII in prompts/responses before shipping traces.

## Main

How to use it without over-instrumenting:

1. **Pick one “golden path” flow** (a user question → tool calls → final answer).
2. Add tracing at the boundaries: request in, model call out, tool call out, response back.
3. Record **inputs/outputs + timings** first. Only add extra metadata (user IDs, tags, datasets) after the baseline works.
4. Create a simple “regression dashboard”: slowest traces, highest error rate, and largest prompt payloads.

The fastest win is spotting which step burns tokens (retrieval, tool results, or prompt templates) and then trimming that step only.

### FAQ

**Q: Do I need a specific model/provider?**
A: No—README says it works with any LLM or framework; focus on consistent trace context instead of vendor-specific fields.

**Q: Should I log full prompts?**
A: Only if allowed. Prefer redaction + sampling for sensitive environments; keep enough context to reproduce failures.

**Q: What breaks during upgrades?**
A: SDK major rewrites can change event shapes. Follow the v4 migration guide before upgrading production services.

## Source & Thanks

> Source: https://github.com/langfuse/langfuse-python
> License: MIT
> GitHub stars: 399 · forks: 266

---

<!-- ZH -->

## 快速使用

```bash
pip install langfuse
```

随后按官方 SDK 文档接入（README 提示 v4 于 2026 年 3 月重写发布）。

## 简介

Langfuse Python SDK 用 decorators 或底层事件把你的 LLM 应用接入可观测性：记录 traces、延迟、token 成本、prompt/response，上线后也能快速定位回归。

- **适合谁：** 需要把 prompt、工具与多供应商调用统一纳入追踪的 Python LLM 应用
- **可搭配：** Python；支持 decorators 或底层事件；可配任意 LLM/provider（见 README）
- **准备时间：** 5–20 分钟

## 实战建议

- README 提示：SDK 在 2026 年 3 月发布 v4 重写版本，升级前先看迁移指南。
- 先从一个关键入口（chat/tool router）开始埋点，再扩到工具调用与后台任务。
- 只记录你能长期保留的内容：把 secrets/PII 在入库前做脱敏或过滤。

## 主要内容

避免“过度埋点”的做法：

1. **先选一条黄金路径**（用户问题 → 工具调用 → 最终回答）。
2. 只在边界处埋点：请求进入、模型调用、工具调用、响应返回。
3. 先记录 **输入/输出 + 耗时**，基础稳定后再逐步加用户标签、数据集等元信息。
4. 做一个最小回归看板：最慢 traces、最高错误率、最大 prompt 体积。

最快收益通常来自定位哪一步最“吃 token”（检索、工具结果、prompt 模板），然后只优化那一步。

### FAQ

**必须绑定某个模型/供应商吗？**
答：不需要。README 表示可配任意 LLM/框架；更重要的是把 trace context 做一致。

**要不要记录完整 prompt？**
答：看合规。建议对敏感字段做脱敏 + 抽样；保留能复现问题的最小上下文即可。

**升级最容易踩什么坑？**
答：大版本重写可能改事件结构；上线前按 v4 迁移指南做对照验证。

## 来源与感谢

> Source: https://github.com/langfuse/langfuse-python
> License: MIT
> GitHub stars: 399 · forks: 266


---
Source: https://tokrepo.com/en/workflows/langfuse-python-sdk-trace-llm-apps
Author: Agent Toolkit