UniLLM Python Client Library

Overview

UniLLM is a unified Python client for accessing multiple large language model (LLM) providers (OpenAI, Anthropic, Gemini, Mistral, Cohere, and more) through a single, easy-to-use interface.
It abstracts away provider-specific details, allowing you to switch models and providers with minimal code changes.

---

Features

• Unified API: One interface for all supported LLM providers and models.
• Drop-in replacements: Use from unillm import openai or from unillm import anthropic_dropin as anthropic for seamless migration from official SDKs.
• Provider/model switching: Change models/providers by changing a string.
• API key management: Pass your UniLLM API key directly or via environment variable.
• Custom error handling: Catch and handle errors with meaningful exceptions.
• Health check: Programmatically check if your backend is up.
• Extensible: Add new providers via adapters.

---

Installation

pip install unillm

---

Quick Start

1. Unified API Example

from unillm import UniLLM

client = UniLLM(api_key="YOUR_API_KEY", base_url="https://your-backend.url")
response = client.chat(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello, world!"}],
    temperature=0.7,
    max_tokens=100
)
print(response.content)

---

2. OpenAI Drop-in Replacement

from unillm import openai
openai.api_key = "YOUR_API_KEY"
openai.api_base = "https://your-backend.url"
response = openai.ChatCompletion.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello!"}],
    temperature=0.7,
    max_tokens=100
)
print(response["choices"][0]["message"]["content"])

---

3. Anthropic Drop-in Replacement

from unillm import anthropic_dropin as anthropic
anthropic.api_key = "YOUR_API_KEY"
anthropic.api_base = "https://your-backend.url"
response = anthropic.ChatCompletion.create(
    model="claude-3-5-sonnet-20240620",
    messages=[{"role": "user", "content": "Hello, Claude!"}],
    temperature=0.7,
    max_tokens=100
)
print(response["choices"][0]["message"]["content"])

---

4. Switching Providers/Models

Just change the model string:

response = client.chat(
    model="claude-3-5-sonnet-20240620",  # Anthropic model
    messages=[{"role": "user", "content": "Hello, Claude!"}]
)

---

5. Migration Notes

• For OpenAI: Change import openai to from unillm import openai and set api_base/api_key.
• For Anthropic: Change import anthropic to from unillm import anthropic_dropin as anthropic and set api_base/api_key.
• All other code can remain the same for chat completions.

---

API Reference

UniLLM Class

UniLLM(api_key=None, base_url="http://localhost:8000")

• api_key: Your UniLLM API key (or set UNILLM_API_KEY env var)
• base_url: URL of your UniLLM backend

chat()

chat(
    model: str,
    messages: List[Dict[str, str]],
    temperature: Optional[float] = None,
    max_tokens: Optional[int] = None,
    stream: bool = False,
    **kwargs
) -> ChatResponse

• model: Model name (e.g., "gpt-4", "claude-3-5-sonnet-20240620")
• messages: List of messages ({"role": ..., "content": ...})
• temperature, max_tokens, etc.: Standard LLM parameters

health_check()

health_check() -> bool

Returns True if backend is healthy.

---

ChatCompletion Drop-ins

• openai.ChatCompletion.create(...)
• anthropic.ChatCompletion.create(...)

---

Supported Models & Providers

• OpenAI: "gpt-4", "gpt-3.5-turbo", etc.
• Anthropic: "claude-3-5-sonnet-20240620", etc.
• Gemini, Mistral, Cohere: (see full list in your backend/docs)

---

Troubleshooting

• 502/connection errors: Check your backend URL and health.
• Model not found: Ensure the model name is supported by your backend.
• API key errors: Make sure your key is valid and set correctly.

---

Contributing

• PRs and issues welcome!
• See [CONTRIBUTING.md] for guidelines.

---

License

MIT