With the Python package any-llm, Mozilla is releasing a unified API for many LLMs in version 1, which is already intended to be stable for production use. This relieves developers when using the models, as they no longer have to maintain a separate adapter for each individual LLM.

The connected models can be in the cloud or local and can be easily switched via the asynchronous API. To improve performance, client connections are reusable. The tool also provides for reasoning a standardized output. Furthermore, any-llm informs users when API modes or endpoints change.

An optional LLM gateway serves for budget and key management and is multi-tenant. Thus, it can serve as an LLM interface for companies.

The List of connected providers on the GitHub page is already long and includes Anthropic, Azure, Databricks, Deepseek, Gemini, Groq, Hugging Face, Llama, Mistral, Ollama, Perplexity, Watsonx, and others. It also contains a table showing which functions any-llm supports for each: Response, Reasoning, Image, and so on.

On the GitHub page of any-llm, you can find a table with supported models and properties.

To use the tool, Python 3.11 and the respective API keys are required, which the tool stores in an environment variable. Mozilla has planned a batch function for the next versions, as well as the integration of further LLMs and additional libraries such as the MCP daemon.

(who)

Don’t miss any news – follow us on Facebook, LinkedIn or Mastodon.

This article was originally published in German. It was translated with technical assistance and editorially reviewed before publication.