Environment:Arize ai Phoenix LLM Provider SDKs
| Knowledge Sources | |
|---|---|
| Domains | LLMs, Evaluation |
| Last Updated | 2026-02-14 06:00 GMT |
Overview
Optional LLM provider SDK dependencies for running evaluations with OpenAI, Anthropic, Google GenAI, MistralAI, LiteLLM, and AWS Bedrock.
Description
This environment provides the optional LLM provider SDK packages needed for the LLM Evaluation Pipeline and Experiment workflows. These are not installed by default with `arize-phoenix-evals`; they must be installed separately depending on which LLM provider is used.
The phoenix-evals package provides a unified `LLM` wrapper class and adapter system that supports multiple providers through a registry pattern. Each provider has its own adapter implementation that translates between the Phoenix evaluation protocol and the provider-specific SDK.
Usage
Use this environment when running LLM-as-judge evaluations or experiments with LLM tasks. Install only the provider SDK(s) you need. The `LLM` class auto-detects which adapters are available based on installed packages.
System Requirements
| Category | Requirement | Notes |
|---|---|---|
| Python | >= 3.8 | phoenix-evals supports Python 3.8+ |
| Network | HTTPS outbound | Required for API calls to LLM providers |
Dependencies
Provider SDKs (install only what you need)
- `openai` >= 1.0.0 (for container: >= 2.14.0)
- `anthropic` >= 0.18.0 (for container: >= 0.78.0)
- `google-genai` >= 1.0.0 (for container on Python >= 3.10: >= 1.50.0)
- `mistralai` >= 1.0.0
- `litellm` >= 1.28.9
- `boto3` (for AWS Bedrock)
- `vertexai` (for Google Vertex AI)
- `azure-identity` (for Azure OpenAI)
Base Dependencies (installed with phoenix-evals)
- `pandas`
- `tqdm`
- `pydantic` >= 2.0.0
- `pystache`
- `jsonpath_ng`
Credentials
Provider API keys must be set as environment variables:
- `OPENAI_API_KEY`: OpenAI API key
- `ANTHROPIC_API_KEY`: Anthropic API key
- `GOOGLE_API_KEY`: Google GenAI API key
- `MISTRAL_API_KEY`: MistralAI API key
- `AWS_ACCESS_KEY_ID` / `AWS_SECRET_ACCESS_KEY`: AWS Bedrock credentials
- `AZURE_OPENAI_API_KEY`: Azure OpenAI API key
Quick Install
# Install phoenix-evals with OpenAI support
pip install arize-phoenix-evals openai
# Install with Anthropic support
pip install arize-phoenix-evals anthropic
# Install with Google GenAI support
pip install arize-phoenix-evals google-genai
# Install with multiple providers
pip install arize-phoenix-evals openai anthropic google-genai
# Install with LiteLLM (supports multiple providers through unified interface)
pip install arize-phoenix-evals litellm
Code Evidence
Provider adapter registry from `packages/phoenix-evals/src/phoenix/evals/llm/registries.py` (detected at import):
# The registry auto-discovers installed provider SDKs
# and registers their adapters when phoenix.evals is imported
Optional dev dependencies from `packages/phoenix-evals/pyproject.toml:39-46`:
[project.optional-dependencies]
dev = [
"anthropic>0.18.0",
"boto3",
"litellm>=1.28.9",
"openai>=1.0.0",
"vertexai",
"mistralai>=1.0.0",
]
Container pinned versions from `pyproject.toml:187-191`:
"anthropic>=0.78.0",
"google-genai>=1.50.0 ; python_version >= '3.10'",
"openai>=2.14.0",
Common Errors
| Error Message | Cause | Solution |
|---|---|---|
| `No adapter found for provider 'openai'` | openai package not installed | `pip install openai` |
| `OPENAI_API_KEY not set` | Missing API key environment variable | Set the appropriate API key env var |
| `anthropic >= 0.18.0 required` | Outdated anthropic package | `pip install --upgrade anthropic` |
Compatibility Notes
- Google GenAI: The container extra requires `google-genai >= 1.50.0` on Python 3.10+, but uses any version on Python < 3.10.
- LiteLLM: Provides a unified interface to multiple providers. Use it when you want provider flexibility without managing individual SDKs.
- Azure OpenAI: Requires `azure-identity` and `aiohttp` packages in addition to `openai`.