The local Smithers app
One command opens a real app on your machine. Tell the concierge what you want in plain English and it builds the workflow, then watch the run come alive as a graph. Everything stays local.
Why this exists
To see a run, you parsed JSON out of the CLI or signed into a hosted product living in some other repo. And to start a workflow, you had to already know the exact command to type.
Runs were rows of JSON. You reconstructed what happened from fields instead of seeing it.
Starting a workflow meant memorizing its name and flags. New work meant reading source, not asking.
How it works
smithers ui --app builds and serves a local app and pops it open in your browser. No sign-in wall.
Say what you want done. A real model reads your workspace and turns the request into an actual runnable workflow.
The run renders as a graph. Steps light up as they finish, gates pause for your call, and you can open any node.
What's inside
A real LLM concierge that reads your project and proposes, then creates, a workflow you can run.
Every run is a graph. Watch steps advance, inspect a node, or replay from a point in time.
Browse the workspace and its diffs from inside the app, next to the run that produced them.
Switch between projects with the workspace picker. The app follows you to the repo you are in.
One command, no account, nothing leaves your machine.