Helicone¶

What it is¶

Helicone is an open-source LLM observability platform that acts as a proxy between your application and LLM providers (like OpenAI, Anthropic, or Groq). It records all requests, responses, and metadata, providing detailed analytics and debugging tools.

What problem it solves¶

It solves the visibility gap in AI applications. Developers often don't know exactly what prompts are being sent, how much they cost in real-time, or where bottlenecks are occurring. Helicone provides a central dashboard for monitoring costs, latency, and model performance with minimal code changes.

Where it fits in the stack¶

Category: Process & Understanding / LLM Gateway & Observability

Typical use cases¶

Cost Monitoring: Tracking spend across multiple models and environments (dev vs. prod).
Prompt Debugging: Inspecting full request/response cycles to fix edge cases.
Performance Benchmarking: Comparing latency between different providers or model versions.
Caching: Reducing costs and latency by caching frequent LLM responses at the proxy level.

Strengths¶

Low Friction: Implementation usually requires just changing the base_url in your LLM client.
Open Source: Can be self-hosted for maximum privacy and data control.
Real-time Analytics: Dashboard provides instant feedback on throughput and error rates.
Feature Rich: Includes tools for A/B testing, user-level tracking, and custom property logging.

Limitations¶

Proxy Latency: Adds a very small amount of network latency (though often offset by caching).
Dependency: Your application's availability depends on the proxy being up (mitigated by Helicone's high availability).

Getting started¶

Basic Integration (OpenAI Python)¶

import openai

client = openai.OpenAI(
  api_key="your-api-key",
  base_url="https://oai.helicone.ai/v1",
  default_headers={
    "Helicone-Auth": f"Bearer {HELICONE_API_KEY}"
  }
)

response = client.chat.completions.create(
  model="gpt-4o",
  messages=[{"role": "user", "content": "Hello!"}]
)

CLI examples¶

helm compose up¶

Deploys the complete Helicone stack using Helm Compose (for self-hosting):

helm compose up

aws s3 ls¶

List backups or logs stored in S3 (standard infrastructure management for Helicone):

aws s3 ls s3://helicone-backups/

curl (Manual Proxy Test)¶

Test the Helicone proxy connectivity directly via curl:

curl https://oai.helicone.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Helicone-Auth: Bearer $HELICONE_API_KEY" \
  -d '{"model": "gpt-4o", "messages": [{"role": "user", "content": "Hello"}]}'

API examples¶

Python (Logging Custom Properties)¶

import openai

client = openai.OpenAI(
  api_key="OPENAI_API_KEY",
  base_url="https://oai.helicone.ai/v1",
  default_headers={
    "Helicone-Auth": f"Bearer {HELICONE_API_KEY}",
    "Helicone-Property-Session": "session_123",
    "Helicone-Property-App": "demo-app"
  }
)

# Properties are automatically tracked in the Helicone dashboard
response = client.chat.completions.create(
  model="gpt-4o",
  messages=[{"role": "user", "content": "Analyze this data."}]
)

Sources / references¶

Contribution Metadata¶

Last reviewed: 2026-05-09
Confidence: high