Mozilla.ai releases universal interface to LLMs

Mozilla.ai, a company backed by the Mozilla Foundation, has released any-llm v1.0, an open-source Python library that provides a single interface to communicate with different large language model (LLM) providers.

any-llm 1.0 was released November 4 and is available on GitHub. With any-llm, developers can use any model, cloud or local, without rewriting a stack every time. This means less boilerplate code, fewer integration headaches, and more flexibility to pick what works best for the developer, Nathan Brake, machine learning engineer at Mozilla.ai, wrote in a blog post. “We wanted to make it easy for developers to use any large language model without being locked into a single provider,” Brake wrote.

Mozilla.ai initially introduced any-llm on July 24. The 1.0 release has a stable, consistent API surface, async-first APIs, and re-usable client connections for high-throughput and streaming use cases, Brake wrote. Clear deprecation and experimental notices are provided to avoid surprises when API changes may occur.

The any-llm v1.0 release adds the following capabilities:

Improved test coverage for stability and reliability

Responses API support 

A List Models API to programmatically query supported models per provider

Re-usable client connections for better performance

Standardized reasoning output across all models, allowing users to access LLM reasoning results regardless of the provider chosen

Auto-updating of the provider compatibility matrix, which shows which features are supported by which providers

Future plans call for support for native batch completions, support for new providers, and deeper integrations inside of the company’s other “any-suite” libraries including any-guardrail, any-agent, and mcpd.