Lmql

Lmql (Language Model Query Language) is a specialized programming language and runtime designed to make working with large language models (LLMs) more structured, controllable, and efficient. It allows developers to define prompts, constraints, and control flow in a single, unified script instead of manually stitching together API calls and string templates. The primary purpose of Lmql is to turn ad-hoc prompt engineering into a reproducible, debuggable development process for LLM-powered applications.

Lmql introduces a query language that combines Python-like syntax with declarative constraints on model outputs, such as enforcing formats, value ranges, or token-level conditions. It supports multi-step interactions, branching logic, and tool-like behaviors within a single query, enabling complex workflows such as retrieval-augmented generation, multi-turn reasoning, or structured extraction. Built-in support for multiple backends (e.g., OpenAI and open-source models) allows the same logic to be reused across different models. Lmql also offers debugging and visualization tools to inspect token-level decisions, constraint satisfaction, and execution traces.