# Helicone — LLM Observability and Prompt Management

> Open-source LLM observability platform. One-line proxy integration for request logging, cost tracking, caching, rate limiting, and prompt versioning across all providers.

## Install

Save in your project root:

## Quick Use

```python
# Just change the base URL — no SDK needed
from openai import OpenAI

client = OpenAI(
    api_key="sk-...",
    base_url="https://oai.helicone.ai/v1",
    default_headers={"Helicone-Auth": "Bearer hlc-..."},
)

# All calls are now logged and tracked
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello!"}],
)
```

## What is Helicone?

Helicone is an open-source LLM observability platform that works as a proxy between your app and LLM providers. With a one-line base URL change (no SDK needed), you get request logging, cost tracking, latency metrics, caching, rate limiting, and prompt versioning — for any LLM provider.

**Answer-Ready**: Helicone is an open-source LLM observability platform. One-line proxy integration (change base URL, no SDK) for request logging, cost tracking, caching, rate limiting, and prompt versioning across OpenAI, Anthropic, and all providers. 5k+ GitHub stars.

**Best for**: Teams running LLM apps in production who need observability without code changes. **Works with**: OpenAI, Anthropic, Google, Azure, any OpenAI-compatible API. **Setup time**: Under 1 minute.

## Core Features

### 1. Zero-SDK Integration
Just change the base URL:

```python
# OpenAI
client = OpenAI(base_url="https://oai.helicone.ai/v1")

# Anthropic
client = Anthropic(base_url="https://anthropic.helicone.ai")

# Azure OpenAI
client = AzureOpenAI(azure_endpoint="https://oai.helicone.ai")
```

### 2. Request Dashboard
Real-time dashboard showing:
- All requests with input/output
- Latency percentiles (p50, p95, p99)
- Token usage per model
- Cost breakdown per user/feature
- Error rates and patterns
- Geographic distribution

### 3. Cost Tracking

```
Dashboard view:
  Today:     $42.50 (1,250 requests)
  This week: $285.30 (8,700 requests)
  By model:
    gpt-4o:        $180 (40%)
    claude-sonnet:  $85 (30%)
    gpt-4o-mini:    $20 (30%)
```

### 4. Caching

```python
# Enable caching with a header
client = OpenAI(
    base_url="https://oai.helicone.ai/v1",
    default_headers={
        "Helicone-Auth": "Bearer hlc-...",
        "Helicone-Cache-Enabled": "true",
    },
)
# Identical requests return cached results instantly
```

### 5. Rate Limiting

```python
headers = {
    "Helicone-RateLimit-Policy": "10;w=60",  # 10 requests per 60 seconds
}
```

### 6. Custom Properties

```python
headers = {
    "Helicone-Property-User": "user-123",
    "Helicone-Property-Feature": "chat",
    "Helicone-Property-Environment": "production",
}
# Filter and group by these properties in the dashboard
```

### 7. Prompt Versioning

```python
headers = {
    "Helicone-Prompt-Id": "customer-support-v3",
}
# Track performance per prompt version
```

## Self-Hosting

```bash
git clone https://github.com/Helicone/helicone
docker compose up -d
# Dashboard at http://localhost:3000
```

## FAQ

**Q: Does it add latency?**
A: Helicone proxy adds < 50ms. Requests are logged asynchronously.

**Q: Is my data safe?**
A: Self-host for full data control. Cloud version is SOC 2 Type II compliant.

**Q: Can I use it with Anthropic Claude?**
A: Yes, change base URL to `https://anthropic.helicone.ai`.

## Source & Thanks

> Created by [Helicone](https://github.com/Helicone). Licensed under Apache 2.0.
>
> [Helicone/helicone](https://github.com/Helicone/helicone) — 5k+ stars

<!-- ZH -->

## 快速使用

```python
# 只改 base_url，无需 SDK
client = OpenAI(base_url="https://oai.helicone.ai/v1")
```

一行代码启用 LLM 可观测性。

## 什么是 Helicone？

开源 LLM 可观测性平台，代理模式一行集成。请求日志、成本追踪、缓存、限流和提示词版本管理。

**一句话总结**：开源 LLM 可观测性平台，一行代理集成（改 base URL），日志、成本、缓存、限流全覆盖，5k+ GitHub stars。

**适合人群**：生产环境运行 LLM 应用需要无侵入可观测性的团队。

## 核心功能

### 1. 零 SDK 集成
只改 base URL，支持 OpenAI、Anthropic、Azure。

### 2. 实时仪表盘
请求、延迟、成本、错误率一览。

### 3. 缓存和限流
Header 开启，无需代码修改。

### 4. 可自托管
Docker Compose 一键部署。

## 常见问题

**Q: 延迟开销？**
A: 代理增加 < 50ms，异步日志。

**Q: 支持 Claude？**
A: 支持，base URL 改为 `anthropic.helicone.ai`。

## 来源与致谢

> [Helicone/helicone](https://github.com/Helicone/helicone) — 5k+ stars, Apache 2.0

---
Source: https://tokrepo.com/en/workflows/8a35faad-0abb-421b-ae18-e869579db4b4
Author: AI Open Source