Terminal — langstage-cli

The terminal stage: a Claude Code-style CLI that runs any LangGraph CompiledGraph from the command line, with streaming, tool-call rendering, and human-in-the-loop approval. It runs your agent — it isn't a bundled coding agent.

dkedar7/langstage-cli PyPI

Quickstart

pip install langstage-cli

langstage-cli --demo "hello"                 # keyless echo agent
langstage-cli -a my_agent.py:graph           # your agent (interactive)
langstage-cli -a my_agent.py:graph "What can you do?"
langstage-cli -f ./prompt.md                 # read the message from a file
langstage-cli --show-config                  # resolved config + sources

Common flags

Flag	Purpose
-a, --agent	Agent spec (path/to/file.py:graph or module:graph)
--demo	Run the built-in keyless echo agent
-f, --file	Read the input message from a file
--no-interactive	Auto-approve tool calls (scripting)
-v, --verbose	Show node names and extra detail
--show-config	Print resolved configuration and exit

In the interactive loop, /q quits, /c clears history, /h shows help.

Programmatic use

from langstage_cli import stream_graph_updates, prepare_agent_input

for chunk in stream_graph_updates(graph, prepare_agent_input(message="Hello!")):
    if chunk.get("chunk"):
        print(chunk["chunk"], end="")