Setting up HoneyHive in your Python Package Manager

Learn how to properly include HoneyHive in your project’s pyproject.toml file using optional dependency groups for clean, targeted installations.

Overview 

HoneyHive provides optional dependency groups that bundle the SDK with specific LLM provider instrumentors and SDKs. This approach offers:

🎯 Targeted Dependencies: Only install what you need
📦 Automatic Resolution: Correct versions guaranteed to work together
🚀 Zero Configuration: Everything ready after installation
🔄 Easy Switching: Change providers by updating dependency group
📊 Clear Intent: Your pyproject.toml shows exactly which providers you use

Single Provider Integration 

Most Common Pattern - Add one provider:

[project]
name = "my-llm-app"
version = "0.1.0"
dependencies = [
    "honeyhive[openinference-openai]",  # OpenAI + instrumentor + SDK
    "fastapi>=0.100.0",
    "uvicorn>=0.20.0"
]

Available Single Provider Options:

dependencies = [
    "honeyhive[openinference-openai]",        # OpenAI GPT models
    "honeyhive[openinference-anthropic]",     # Anthropic Claude models
    "honeyhive[openinference-google-ai]",     # Google Gemini models
    "honeyhive[openinference-bedrock]",   # AWS Bedrock multi-model
    "honeyhive[openinference-azure-openai]",  # Azure-hosted OpenAI
]

Multiple Provider Integration 

Production Apps with Multiple Providers:

[project]
name = "my-multi-provider-app"
version = "1.0.0"
dependencies = [
    "honeyhive[openinference-openai,openinference-anthropic,openinference-google-ai]",  # Multiple providers
    "fastapi>=0.100.0",
    "pydantic>=2.0.0"
]

Popular Provider Combination:

dependencies = [
    "honeyhive[openinference-llm-providers]",  # OpenAI + Anthropic + Google (most popular)
]

Framework-Specific Integration 

LangChain Applications:

[project]
name = "my-langchain-app"
dependencies = [
    "honeyhive[openinference-langchain]",     # LangChain + instrumentor
    "honeyhive[openai]",        # Add your LLM provider
    "chromadb>=0.4.0"
]

LlamaIndex RAG Applications:

[project]
name = "my-rag-app"
dependencies = [
    "honeyhive[llamaindex]",    # LlamaIndex + instrumentor
    "honeyhive[openai]",        # Add your LLM provider
    "pinecone-client>=2.0.0"
]

DSPy Programming Framework:

[project]
name = "my-dspy-app"
dependencies = [
    "honeyhive[dspy]",          # DSPy + instrumentor
    "honeyhive[openai]",        # Add your LLM provider
]

Optional Dependencies Pattern (Recommended)

Flexible User Choice - Let users pick providers:

[project]
name = "my-flexible-library"
version = "0.1.0"
dependencies = [
    "honeyhive",  # Core SDK only - no provider lock-in
    "pydantic>=2.0.0",
    "httpx>=0.24.0"
]

[project.optional-dependencies]
# Let users choose their providers
openai = ["honeyhive[openinference-openai]"]
anthropic = ["honeyhive[anthropic]"]
google = ["honeyhive[google-ai]"]
aws = ["honeyhive[bedrock]"]
azure = ["honeyhive[azure-openai]"]

# Framework integrations
langchain = ["honeyhive[openinference-langchain]"]
llamaindex = ["honeyhive[llamaindex]"]

# Convenience groups
popular = ["honeyhive[llm-providers]"]        # OpenAI + Anthropic + Google
all-providers = ["honeyhive[all-integrations]"]  # Everything

# Development dependencies
dev = [
    "honeyhive[openai,anthropic]",  # Test with multiple providers
    "pytest>=7.0.0",
    "black>=23.0.0",
    "mypy>=1.0.0"
]

Users can then install with:

# Install your library with OpenAI support
pip install my-flexible-library[openai]

# Install with multiple providers
pip install my-flexible-library[openai,anthropic]

# Install with all providers for testing
pip install my-flexible-library[all-providers]

All Integrations (Kitchen Sink)

Enterprise Apps with Comprehensive Provider Support:

[project]
name = "enterprise-llm-platform"
version = "2.0.0"
dependencies = [
    "honeyhive[all-integrations]",  # All providers + frameworks
    "fastapi>=0.100.0",
    "sqlalchemy>=2.0.0",
    "redis>=4.0.0"
]

Note: Only use all-integrations if you actually need multiple providers. For most apps, specific provider groups are better.

Tool-Specific Examples 

requirements.txt (pip)

# Core app dependencies
honeyhive[openinference-openai,openinference-anthropic]>=1.0.0
fastapi>=0.100.0
uvicorn>=0.20.0

# Framework integration example
# honeyhive[openinference-langchain]>=1.0.0

# Multiple providers
# honeyhive[openinference-llm-providers]>=1.0.0

# Install from requirements.txt
pip install -r requirements.txt

# Or install directly
pip install "honeyhive[openinference-openai,openinference-anthropic]>=1.0.0"

uv

# Initialize new project with uv
uv init my-llm-app
cd my-llm-app

# Add HoneyHive with providers
uv add "honeyhive[openinference-openai]"
uv add "honeyhive[openinference-anthropic]"

# Or add multiple providers at once
uv add "honeyhive[openinference-openai,openinference-anthropic]"

# Add framework integration
uv add "honeyhive[openinference-langchain]"

# Run your application
uv run python main.py

# pyproject.toml (generated by uv)
[project]
name = "my-llm-app"
version = "0.1.0"
dependencies = [
    "honeyhive[openinference-openai,openinference-anthropic]>=1.0.0",
    "fastapi>=0.100.0",
]

Poetry

[tool.poetry.dependencies]
python = "^3.11"
honeyhive = {extras = ["openinference-openai", "openinference-anthropic"], version = "^1.0.0"}
fastapi = "^0.100.0"

pip-tools (requirements.in)

# Core app dependencies
honeyhive[openinference-openai,openinference-anthropic]>=1.0.0
fastapi>=0.100.0
uvicorn>=0.20.0

# Compile to requirements.txt
pip-compile requirements.in

# Install
pip-sync requirements.txt

Pipenv

[packages]
honeyhive = {extras = ["openinference-openai"], version = "*"}
fastapi = "*"

Hatch

[project]
dependencies = [
    "honeyhive[openinference-google-ai]",
]

[project.optional-dependencies]
dev = ["honeyhive[openinference-openai,openinference-anthropic]"]  # More providers for testing

Available Optional Dependencies 

🤖 LLM Providers

Extra	What’s Included
`openai`	OpenAI SDK + OpenInference OpenAI instrumentor
`anthropic`	Anthropic SDK + OpenInference Anthropic instrumentor
`google-ai`	Google Generative AI SDK + OpenInference Google instrumentor
`google-adk`	Google Agent Development Kit + OpenInference ADK instrumentor
`bedrock`	Boto3 + OpenInference Bedrock instrumentor
`azure-openai`	OpenAI SDK + Azure Identity + OpenInference OpenAI instrumentor
`mcp`	OpenInference MCP instrumentor for Model Context Protocol

🔧 Framework Integrations

Extra	What’s Included
`langchain`	LangChain + OpenInference LangChain instrumentor
`llamaindex`	LlamaIndex + OpenInference LlamaIndex instrumentor
`dspy`	DSPy + OpenInference DSPy instrumentor

🌟 Additional Providers

Extra	What’s Included
`cohere`	Cohere SDK + OpenInference Cohere instrumentor
`huggingface`	Transformers + OpenInference HuggingFace instrumentor
`mistralai`	Mistral AI SDK + OpenInference Mistral instrumentor
`groq`	Groq SDK + OpenInference Groq instrumentor
`ollama`	Ollama SDK + OpenInference Ollama instrumentor
`litellm`	LiteLLM + OpenInference LiteLLM instrumentor

📦 Convenience Groups

Extra	What’s Included
`llm-providers`	OpenAI + Anthropic + Google AI (most popular providers)
`all-integrations`	All available instrumentors and SDKs

Best Practices 

✅ Do This

# Good: Specific providers you actually use
dependencies = ["honeyhive[openai,anthropic]"]

# Good: Let users choose in a library
[project.optional-dependencies]
openai = ["honeyhive[openinference-openai]"]

❌ Avoid This

# Avoid: Installing everything when you only use OpenAI
dependencies = ["honeyhive[all-integrations]"]

# Avoid: Manual instrumentor management
dependencies = [
    "honeyhive",
    "openinference-instrumentation-openai",  # Use honeyhive[openinference-openai] instead
    "openai"
]

🎯 Choosing the Right Pattern

Application: Use specific provider extras like honeyhive[openinference-openai]
Library: Use optional dependencies to let users choose
Enterprise: Consider honeyhive[llm-providers] for popular providers
Testing: Use honeyhive[all-integrations] for comprehensive testing

Migration from Manual Installation 

Before (Manual):

dependencies = [
    "honeyhive",
    "openinference-instrumentation-openai",
    "openinference-instrumentation-anthropic",
    "openai",
    "anthropic"
]

After (Optional Dependencies):

dependencies = [
    "honeyhive[openai,anthropic]"  # Much cleaner!
]

Benefits of Migration:

Fewer Dependencies: One line instead of five
Version Compatibility: Guaranteed to work together
Easier Maintenance: Update one package instead of tracking multiple
Clearer Intent: Obvious which providers you use

Troubleshooting 

Import Errors After Installation

Make sure you installed the right extra:

# If using OpenAI
pip install honeyhive[openinference-openai]

# If using multiple providers
pip install honeyhive[openinference-openai,openinference-anthropic]

Version Conflicts

The optional dependencies are curated to avoid conflicts. If you see version conflicts:

Use the optional dependency groups instead of manual installation
Update to the latest HoneyHive version
Check that you’re not manually specifying conflicting versions

Missing Provider Support

If a provider isn’t available as an optional dependency:

# Fall back to manual installation
pip install honeyhive
pip install openinference-instrumentation-<provider>
pip install <provider-sdk>

# Then file an issue to request the provider be added!

Next Steps 

Quick Start: How-to Guides - Choose your provider integration
Examples: Tutorials - See complete examples
Deployment: Production Deployment Guide - Production deployment guides