UI¶

everystep ships a self-contained single-page UI — inline CSS and JS, no CDN assets — that shows workflow runs, live progress, step outputs, and the runners that hold work. Mount it in your application:

# urls.py
from django.urls import include, path

urlpatterns = [
    path("everystep/", include("everystep.urls")),
]

Routes¶

The UI page calls /api/stream for live updates: the server re-sends the full runs snapshot whenever it changes (checked once a second, with a ping every 15 idle seconds to keep proxies from closing the connection). Clients whose proxies do not stream SSE fall back to polling /api/runs.

What you see¶

Runs — the 200 most recent, newest first, each with its name, status (scheduled, running, completed, failed, stopped, blocked), the claiming runner, and a step progress count. Runners are the distinct claimed_by values of running workflows with their in-flight counts; an idle runner has nothing in flight and does not appear.

One run — the run's arguments, result or error, and its step graph with every recorded step: id, function, args, kwargs, result, error.

Any run can be opened directly by id: paste it into the input in the header, or load the page with a #run/<id> fragment — including runs older than the 200 shown in the list.

The step graph¶

For a workflow whose body is a straight sequence of steps and parallel forks, the UI renders the intended structure of the workflow — parsed from the source via the AST, reproducing the exact step ids the runtime assigns — as a top-to-bottom DAG in the style of Argo Workflows: each step is a node, each parallel a fan-out / fan-in, with the step's live status as the node color:

• recorded steps show their outcome (done, failed);
• a step started by an unsafe-to-repeat mark but never recorded shows uncertain — its node card carries the everystep_resolve_step command that resolves the run;
• the in-flight step of a running workflow is shown in flight (pulsing);
• the rest are pending.

Zoom with the mouse wheel, pan by dragging, refit with the + / − / fit buttons. Clicking a node opens the step's id, function, args, kwargs, result, and error below the canvas.

Bodies the parser cannot prove — control flow, loops, dynamic step ids, calls through local names — fall back to a flat view built from the recorded step ids alone, drawn through the same layout, with the reason for the fallback shown. The parser inlines plain (non-step) function calls, so helper functions that just call steps are still drawn in place.

Security¶

Like the metrics endpoint, the UI is unauthenticated: it exposes run arguments, results, and errors. Keep the mount behind network segmentation or your own authentication.