CLI reference¶
The langstitch command scaffolds, inspects, compiles, and runs projects.
langstitch new <name> [--dir PATH] [--force] scaffold an Agentic Development project
langstitch plugin new <name> [--dir PATH] [--force] scaffold a Plugin Creator pack project
langstitch pack [--root PATH] [--out DIR] [--document PATH] [--platforms LIST] build a .langstitch-pack.zip
langstitch validate-pack [--root PATH] [--document PATH] validate pack manifests + IR
langstitch info [--root PATH] load config + list components
langstitch run [--root PATH] [--host] [--port] start the API server
langstitch compile [document.langstitch.json] [--out DIR] [--force] IR v2 -> Python project (or config precompile)
langstitch get <json.path> [--root PATH] resolve a config path, print as JSON
langstitch version print the SDK version
Install the compiler extra for IR documents: pip install "langstitch-sdk[compiler]".
Commands¶
langstitch new¶
Scaffold a new Agentic Development project with the conventional project structure.
langstitch plugin new¶
Scaffold a Plugin Creator pack project (langstitch.project.json with kind: plugin-creator, plus pack.json and IR).
langstitch pack¶
Build a multi-platform .langstitch-pack.zip from a Plugin Creator project. Validation is fail-closed by default — the command exits with errors if manifests, IR, or platform artifacts are invalid or missing. Pass --allow-partial to emit a zip when some platform builds are absent.
langstitch validate-pack¶
Validate pack manifests, IR, and per-platform templates without writing a zip. Useful before marketplace publish.
langstitch info¶
Load configuration and list every registered component (nodes, graphs, tools, agents, guardrails, policies, personas, MCP primitives).
langstitch run¶
Start the FastAPI server (requires the server extra). Exposes /health,
/info, /invoke.
langstitch compile¶
With a document — compile a LangTailor IR v2 *.langstitch.json into a Python
project (requires the compiler extra). See IR v2 compiler.
Without a document — precompile application.yaml → application.json for fast
startup. The JSON file, when present, loads directly and takes precedence.
langstitch get¶
Resolve a config path with the JSON-path syntax and print the result as JSON.