multi-llm-provider 0.1.0

Unified multi-provider LLM interface supporting Claude, OpenAI, Mistral, Ollama, and Clipboard

multi-llm-provider

Unified multi-provider LLM interface supporting Claude, OpenAI, Mistral, Ollama, and Clipboard.

Installation

pip install multi-llm-provider

Optional providers

Install only the providers you need:

pip install multi-llm-provider[claude]    # Anthropic Claude
pip install multi-llm-provider[openai]    # OpenAI GPT
pip install multi-llm-provider[mistral]   # Mistral AI
pip install multi-llm-provider[ollama]    # Ollama (local)
pip install multi-llm-provider[all]       # All providers

Quick start

from multi_llm_provider import AIProvider, AIProviderFactory

# Create an analyzer for Claude
analyzer = AIProviderFactory.create(
    provider=AIProvider.CLAUDE_API,
    model="claude-sonnet-4-20250514",
    system_prompt="You are a helpful assistant.",
)

# Analyze content
result = analyzer.analyze("Explain quantum computing in simple terms.")
print(result)

Supported providers

Provider	Enum value	Required extra
Anthropic Claude	AIProvider.CLAUDE_API	claude
OpenAI GPT	AIProvider.OPENAI_API	openai
Mistral AI	AIProvider.MISTRAL_API	mistral
Ollama (local)	AIProvider.OLLAMA	ollama
Clipboard (manual)	AIProvider.CLIPBOARD	none

Architecture

All providers implement the AIAnalyzer abstract base class:

from multi_llm_provider import AIAnalyzer

class AIAnalyzer(ABC):
    @abstractmethod
    def analyze(self, content: str) -> str:
        """Send content to the LLM and return the response."""
        ...

The AIProviderFactory creates the appropriate analyzer based on the provider enum:

analyzer = AIProviderFactory.create(
    provider=AIProvider.OPENAI_API,
    model="gpt-4o",
    system_prompt="You are a data analyst.",
    temperature=0.7,
    max_tokens=4096,
)

Clipboard provider

The ClipboardAnalyzer copies content to the system clipboard with instructions, then waits for the user to paste the LLM response back. Useful for manual workflows or when API access is not available.

Error handling

from multi_llm_provider import ConfigError, WorkflowError

try:
    result = analyzer.analyze("some content")
except WorkflowError as e:
    print(f"LLM call failed: {e}")
except ConfigError as e:
    print(f"Configuration error: {e}")

Development

git clone https://github.com/stephanejouve/multi-llm-provider.git
cd multi-llm-provider
poetry install --with dev,extras
poetry run pytest tests/ -v
poetry run black src/ tests/ --check --line-length=100
poetry run ruff check src/
poetry run isort src/ tests/ --check-only --profile black --line-length=100

License

MIT