Kenneth Wolters offers "Litelm," a library specifically designed to route LLM (Large Language Model) calls across different providers and convert message formats.
Litelm: A Lightweight Version of LiteLLM's Core Functionality
This article is a translation. Read the Japanese original
Litelm extracts only the core pathways from LiteLLM—which features multi-functional proxy server and caching capabilities—such as model routing, message conversion, streaming, tool use, and embeddings. The source code is kept to approximately 2,900 lines, and dependencies are limited to just 2 from openai and httpx.
The API interface inherits from LiteLLM, maintaining the same function names, arguments, and response formats. This design allows existing LiteLLM users to migrate simply by updating their import statements. Additionally, it supports routing to 19 providers and is compatible with any OpenAI-compatible endpoint.
It is reported that much of the code was developed using Claude Code and Pi, and it is currently available as an alpha version.
Sources
- Litelm: LiteLLM Without the Bloat （Hacker News Frontpage、2026-09-11）