Simon Willison released version 0.1a0 of llm-chat-completions-server, a tool that exposes his local LLM collection through an OpenAI Chat Completions compatible endpoint. The software runs on a localhost server and allows standard API requests where each incoming message extends a previous conversation history. This approach differs from the default LLM workflow where the client manages conversation state and requests grow progressively longer. The new plugin uses a schema design based on hashes of individual message parts to de-duplicate these extended contexts efficiently. GPT-5.6 Sol generated the code for the project, demonstrating a strong understanding of the specific API shape required for compatibility.
The release addresses a practical need for developers who prefer the familiar ChatGPT Completions interface while running models locally. By mapping local models to this standard format, the tool simplifies integration with existing workflows that rely on stateful chat histories. It also provides a testing ground for the content-addressable log system introduced in the recent LLM 0.32rc1 update.
* Runs on port 9001 by default
* Supports models from any installed plugins
* Generated entirely by GPT-5.6 Sol
