llm-chat-completions-server 0.1a0 has been released. The tool provides a local HTTP endpoint compatible with the OpenAI Chat Completions API shape, exposing the LLM models available through any llm plugins you have installed. It serves incoming message sequences as /v1/chat/completions requests in a ChatGPT-style conversation flow.
Why this matters
One of the goals of the content-addressable logs introduced in LLM 0.32rc1 was to support OpenAI Chat Completion–style requests, where each incoming message extends the previous conversation. In such usage the client typically sends increasingly long message arrays (for example: a user question, the assistant reply, and a follow-up user question). The new LLM schema is designed to deduplicate these expanding conversation states by hashing individual message parts.
How to try it
A plugin was built to test this functionality and can be installed and run with these commands:
- uv tool install llm --pre
- llm install llm-chat-completions-server
- llm chat-completions-server -p 9001
The last command starts a localhost server on port 9001 that exposes all LLM models provided by installed llm plugins via a ChatGPT Completions–compatible endpoint.
API behavior
The announcement includes a typical OpenAI Chat Completions–style example request:
curl http://localhost:8002/v1/chat/completions
-H 'Content-Type: application/json'
-d '{
"model": "qwen3.5-4b",
"messages": [
{"role": "user", "content": "Capital of France?"},
{"role": "assistant", "content": "Paris."},
{"role": "user", "content": "Germany?"}
]
}'
Because the conversation state is tracked by the client, each request typically grows longer; the LLM schema deduplicates repetitive data using hashes of message components.
Note
The post also states that “GPT-5.6 Sol wrote the whole thing,” noting that GPT-5.6 Sol knows the OpenAI Chat Completions API shape very well.
In short: llm-chat-completions-server 0.1a0 offers a straightforward way to serve locally hosted LLMs through a Chat Completions–compatible API while leveraging LLM 0.32rc1's content-addressable logging to reduce duplication in conversation histories.
