Environment:BerriAI Litellm Python Runtime

Knowledge Sources	BerriAI/litellm pyproject.toml
Domains	Infrastructure, LLM_Gateway
Last Updated	2026-02-15 16:00 GMT

Overview

Python 3.9+ runtime environment with core dependencies for the LiteLLM SDK library (httpx, openai, tiktoken, pydantic).

Description

This environment defines the minimum Python runtime and core package dependencies required to use the LiteLLM SDK for making LLM API calls. It covers the base litellm package without the optional proxy server or caching extras. The SDK supports Python 3.9 through 3.13 (and experimentally 3.14 for grpcio), running on Linux, macOS, and Windows.

Usage

Use this environment for any project that calls litellm.completion(), litellm.embedding(), or other SDK-level functions. This is the minimum prerequisite for all LiteLLM functionality. The proxy server, caching backends, and observability integrations require additional optional dependencies.

System Requirements

Category	Requirement	Notes
OS	Linux, macOS, Windows	All platforms supported; some optional deps (uvloop, pyroscope-io) are Linux/macOS only
Python	>= 3.9, < 4.0	3.11 recommended for Docker deployments
Disk	200MB	For package installation and tiktoken cache

Dependencies

System Packages

`python3` >= 3.9
`pip` (package installer)

Python Packages (Core)

`openai` >= 2.8.0
`httpx` >= 0.23.0
`tiktoken` >= 0.7.0
`pydantic` >= 2.5.0, < 3.0.0
`aiohttp` >= 3.10
`jinja2` >= 3.1.2, < 4.0.0
`jsonschema` >= 4.23.0, < 5.0.0
`importlib-metadata` >= 6.8.0
`python-dotenv` >= 0.2.0
`fastuuid` >= 0.13.0
`tokenizers` (any version)
`click` (any version)

Credentials

The following environment variables are used for provider API access:

`OPENAI_API_KEY`: OpenAI API key for GPT models.
`ANTHROPIC_API_KEY`: Anthropic API key for Claude models.
`LITELLM_MODE`: Set to "DEV" or "PRODUCTION" to control runtime behavior.
`LITELLM_DROP_PARAMS`: Boolean flag to silently drop unsupported parameters.
`LITELLM_MODEL_COST_MAP_URL`: Custom URL for model cost map (optional override).

Quick Install

# Install core LiteLLM SDK
pip install litellm

# Or with proxy server extras
pip install 'litellm[proxy]'

# Or with all extras for full deployment
pip install 'litellm[proxy,extra_proxy,caching]'

Code Evidence

Python version constraint from `pyproject.toml:22`:

python = ">=3.9,<4.0"

Core dependency declarations from `pyproject.toml:23-34`:

fastuuid = ">=0.13.0"
httpx = ">=0.23.0"
openai = ">=2.8.0"
python-dotenv = ">=0.2.0"
tiktoken = ">=0.7.0"
importlib-metadata = ">=6.8.0"
tokenizers = "*"
click = "*"
jinja2 = "^3.1.2"
aiohttp = ">=3.10"
pydantic = "^2.5.0"
jsonschema = ">=4.23.0,<5.0.0"

Lazy loading system to avoid heavy imports at startup from `litellm/__init__.py:1605`:

if os.getenv("LITELLM_DISABLE_LAZY_LOADING"):
    # Load all modules eagerly
    ...

Common Errors

Error Message	Cause	Solution
`ImportError: No module named 'openai'`	Core dependency not installed	`pip install litellm`
`pydantic.errors.PydanticImportError`	Pydantic v1 installed instead of v2	`pip install 'pydantic>=2.5.0'`
`ImportError: cannot import name 'Version'`	Old importlib-metadata	`pip install 'importlib-metadata>=6.8.0'`

Compatibility Notes

Windows: `uvloop` and `pyroscope-io` are not available. The proxy server falls back to asyncio event loop.
Python 3.14: Requires `grpcio >= 1.75.0` due to wheel availability. Versions 1.68.x-1.73.0 have known bugs.
Python 3.9: Some optional packages like `mcp`, `a2a-sdk`, `polars`, and `mlflow` require Python >= 3.10.

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment