Wrapper4AI

Multi-provider, pluggable LLM wrapper with token counting, history management, streaming, and seamless extensibility.

Unified Python client for OpenAI, Anthropic, Google Gemini, Mistral AI, DeepSeek, AWS Bedrock, Hugging Face, and Perplexity AI.

Features

• Unified interface for multiple LLM providers
• Chat history with optional tracking and token trimming
• Streaming support for all providers
• Token counting (tiktoken when available)
• Provider registry — register custom backends via BaseProvider
• Config from env — API keys and regions from environment variables

Installation

pip install wrapper4ai

Install with optional dependencies for the providers you use:

# One provider
pip install wrapper4ai[openai]

# Multiple
pip install wrapper4ai[openai,anthropic,google,mistral]

# All providers
pip install wrapper4ai[all]

Optional extras: openai, anthropic, google, mistral, bedrock, huggingface, all, dev.

Supported Providers

Usage

Connect and chat

from wrapper4ai import connect

client = connect("openai", "gpt-4o", api_key="sk-...")
response = client.chat("Tell me a joke.")
print(response)

API keys can be set via environment variables (e.g. OPENAI_API_KEY, ANTHROPIC_API_KEY, GOOGLE_API_KEY, MISTRAL_API_KEY) or passed as api_key=.

Streaming

client = connect("anthropic", "claude-sonnet-4-20250514")
for chunk in client.stream("Write a short poem."):
    print(chunk, end="")

With history

History is kept by default; you can clear it or turn it off per call:

client = connect("openai", "gpt-4o")
client.chat("My name is Alex.")
client.chat("What is my name?")  # uses history
client.chat("Hello!", use_history=False)  # no history update
client.clear_history()

Stateless completion

messages = [
    {"role": "user", "content": "What is 2+2?"},
]
reply = client.complete(messages)  # does not change client history

System prompt and token counting

client = connect("openai", "gpt-4o", system_prompt="You are a helpful assistant.")
client.chat("Hi!")
n = client.count_tokens("Some text")
n = client.count_tokens(client.history)
total = client.total_tokens_used
title = client.generate_title("A long discussion about AI")

List providers and custom backends

from wrapper4ai import list_providers, connect
from wrapper4ai.providers import BaseProvider, register_provider

print(list_providers())  # ['anthropic', 'bedrock', 'deepseek', 'google', ...]

class MyProvider(BaseProvider):
    def generate(self, messages): ...
    def stream(self, messages): ...
    def count_tokens(self, messages): ...

register_provider("my_backend", MyProvider)
client = connect("my_backend", "my-model")

Testing

pip install -e ".[dev]"
PYTHONPATH=. pytest tests/ -v

Tests use a mock provider; no API keys required for the test suite.

License

MIT.