llm://
A name system for language models.
Every tool writes model names differently, and none of them say where the model lives, which protocol it speaks, or how to authenticate. llmns puts the whole address in one portable string: the way a URL names a page, a reference names a model.
01Location. A DNS hostname or an IP, exactly as URLs have always said where. Nothing to look up that DNS can't answer.
02Transport. llm, llms, llms+grpc: the scheme states TLS and protocol, mirroring http and https.
03State. A typed pin (hash, version, or name) declares exactly which state of the model, and how firmly it is fixed.
04Credential. A name in your local store, resolved by your client. The reference never carries the secret.
llms://api.anthropic.com/claude-fable-5
llms://api.openai.com/gpt-5@version:2026-03-01
llm://localhost:11434/llama3.2:3b?api=openai
llms+grpc://triton.internal:8001/qwen3-ft@name:step-2000