CLI Overview

pdsa coaches an AI agent through Deming’s cycle and accumulates each step into a per-project graph DB. Run it with no arguments for full help; add --help to any command for its usage.

pdsa            # full help
pdsa <command> --help

Commands

Command	Purpose
plan	Enter a plan → LLM sets a verifiable expected evaluation (starts a cycle)
do	Report what you did → organizes Plan → Do
study	Report results → learning + verdict (met/partial/unmet)
act	Learnings + reinforcement decision (closes the cycle)
status	Recent cycles + expectation hit-rate
eval	Per-cycle expected / verdict / actual + hit-rate
recall	Read back prior-cycle learnings (planning context)
project	Set / list / show / clear the active project
view	Local graph viewer
config	LLM key / model / provider / auth / language
check	Verify the LLM with a real round-trip
models	List models the endpoint serves
init	Install the PDSA skill into a workspace
guide	One-off PDSA advice from the LLM
run	Run the demo PDSA feedback cycle (Akka.Streams)
update	Check the latest version and update (npm global)
version	Version + runtime + stack

Global options

Option	Effect
--lang <en|ko|auto>	Language for this invocation. See Language.
--project <name>	Run this command against a specific project’s DB (per-command).
--json	Machine-readable JSON instead of prose. Opt-in on plan/do/study/act/status/eval/recall. See For AI Agents.
--full	On status/eval, print prose without the 70/90-char truncation.
--help	Usage for the command.

Where state lives

Global config : {LocalAppData}/pdsa-cli/openai.json
Graph DBs      : {LocalAppData}/pdsa-cli/{project}/graph.kuzu
Repo secret    : .secret/openai.json   (optional, lowest LLM precedence)

Text is safe

Quotes and newlines inside your Plan/Do/Study text are stored via parameter binding — pass them as-is.