Alternative Integration Methods

Add Helicone observability to your existing LLM setup without changing your SDK or architecture

These integration methods let you add Helicone’s observability to your existing LLM applications without adopting the AI Gateway. Perfect for teams with established codebases or specific provider requirements.

When to Use These Methods

Existing Codebase

Your application is already built with provider-specific SDKs and refactoring would be costly

Provider Features

You need native SDK features like OpenAI’s complex function calling or Anthropic’s specific formats

Minimal Changes

You want observability without changing your current authentication or request flow

Framework Users

You’re using LangChain, LlamaIndex, or other frameworks that handle LLM calls

Quick Comparison

Aspect	Alternative Integrations	AI Gateway
Setup	Change endpoint URL only	✅ Simple - one endpoint for 100+ models
Code changes	Minimal - keep existing SDK	Minimal - unified OpenAI format
Switch providers	❌ Rewrite for each provider	✅ Just change model name
Provider features	✅ Full native support	Standard OpenAI format
Observability	✅ Full Helicone features	✅ Full Helicone features
Fallbacks	❌ Manual implementation	✅ Automatic
Best for	Existing apps, native features	New projects, multi-provider

Direct Provider Integrations

Popular Providers

Simply change your base URL to add Helicone observability:

OpenAI

Proxy: oai.helicone.ai
Keep using OpenAI SDK as-is

Anthropic

Proxy: anthropic.helicone.ai
Works with Claude models

Google Gemini

Proxy: Available
Support for Gemini models

Together AI

Headers: Add to existing
Open source model hosting

OpenRouter

Proxy: openrouter.helicone.com
Access 100+ models with one API

Cloud Providers

AWS Bedrock

Use AWS SDK with Helicone logging for Bedrock models

Azure OpenAI

OpenAI SDK with Azure endpoints and Helicone headers

Google Vertex AI

Vertex AI with Helicone observability

Speed-Optimized Providers

Groq

Ultra-fast inference with LPU technology

Fireworks AI

Fast open-source model serving

Perplexity

Search-optimized language models

Framework Integrations

If you’re using an AI framework, add Helicone with minimal configuration:

LangChain

Monitor chains, agents, and retrieval flows

LlamaIndex

Track document queries and RAG pipelines

LiteLLM

Unified interface with 100+ models

Vercel AI SDK

Stream responses in Next.js apps

CrewAI

Monitor multi-agent workflows

Custom Integration Options

Async Logging

For zero-latency observability, log requests asynchronously after they complete:

Async Logging

Send logs to Helicone after receiving LLM responses - no proxy latency

Custom HTTP Clients

Any HTTP client can work with Helicone:

cURL

curl https://oai.helicone.ai/v1/chat/completions \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Helicone-Auth: Bearer $HELICONE_API_KEY" \
  -d '{"model": "gpt-4", "messages": [...]}'

Implementation Patterns

Proxy Pattern

Change your base URL to route through Helicone:

• OpenAI: api.openai.com → oai.helicone.ai
• Anthropic: api.anthropic.com → anthropic.helicone.ai
• Add Helicone-Auth header

Header Pattern

Keep your existing endpoint, add Helicone headers:

• Add Helicone-Auth header
• Add custom properties as needed
• Works with any provider

Async Pattern

Log after the request completes:

• Make normal LLM request
• Send logs to Helicone async endpoint
• Zero added latency

Still Considering the AI Gateway?

The AI Gateway might be better if you:

• Want automatic fallbacks between providers
• Need a unified API for multiple models
• Are starting a new project
• Want built-in prompt management

Learn About AI Gateway

See how the AI Gateway simplifies multi-provider LLM applications