Shire is described as a simple AI coding agent language that lets a large language model (LLM) converse freely with, and control, an integrated development environment (IDE) in order to automate programming. In Shire, how information from the IDE is handled and how the IDE interacts with remote agents are defined the way a program is written.
It is presented in 云端与 IDE 智能体整合：解决工具碎片化，实现 AI 全流程自动编码 as an implementation of collaboration between cloud agents and IDE agents, in which an agent orchestration system on the IDE side calls agents running in the cloud as well as local ones.
Details
The example given in that post is a file that begins with a metadata header — a name, variables and an
after-streaming action — followed by the prompt text. One variable is filled by a thread function that runs a
shell script calling a remote agent deployed on the Dify platform and extracts the answer with jsonpath; the
prompt then combines that requirement with the user’s database information and tables. After the model has
analysed the requirement, an execute function calls a second, local Shire agent (gen-sql.shire) that holds
basic SQL conventions, so that the generated SQL follows enterprise standards. Further examples are published in
the shire-lang/shire-spring-java-demo GitHub repository.