Web Dashboard

The graflag gui command starts a web dashboard for monitoring and managing experiments.

Usage

graflag gui [--host HOST] [--port PORT] [--debug] [--config FILE]

Default: http://0.0.0.0:5000

Warning

0.0.0.0 binds every interface, so the dashboard is reachable from any host that can route to the machine. The dashboard has no authentication, and every action it exposes – running experiments, deleting experiment directories, reading files – is carried out on the swarm manager as root. Run it on a trusted network, or bind it explicitly:

graflag gui --host 127.0.0.1

Experiment, method and dataset names arriving over HTTP are validated against ^[A-Za-z0-9._-]{1,200}$ before they reach any remote command, but that guard is not a substitute for network isolation.

Features

• Cluster status – connection and Swarm health in the navbar

• Methods and datasets – browse available resources with pagination

• Run experiments – select method, dataset, parameters, and deploy

• Live updates – the experiment and service lists are pushed over a WebSocket (update events) as the background poller notices changes

• Logs – fetched over HTTP from /api/experiments/<name>/logs?tail=200: every two seconds while the output keeps changing, backing off to thirty seconds when it goes quiet, and not at all once the run has finished. They are polled, not streamed, and only the last 200 lines are shown (the last 1000 in the Run form’s live view); use graflag logs -e EXP -f for a true stream

• Evaluation viewer – the scalar metrics (counts as integers, with the number of samples the evaluator excluded when there are any) and every plot in a grid, each opening full size in a lightbox

• Notifications – browser notifications for experiment status changes

• Dark/light theme

API Endpoints

Method	Endpoint	Description
GET	/api/cluster/info	Cluster status
GET	/api/methods	List methods
GET	/api/datasets	List datasets
GET	/api/experiments	List experiments
GET	/api/experiments/<name>/results	Experiment results
GET	/api/experiments/<name>/evaluation	Evaluation metrics
GET	/api/experiments/<name>	Experiment details
GET	/api/experiments/<name>/logs	Experiment logs (last tail lines, default 100)
GET	/api/experiments/<name>/plot/<plot_name>	Evaluation plot image
GET	/api/methods/<method_name>	Method details
GET	/api/services	Running services
POST	/api/run	Run experiment
POST	/api/experiments/<name>/evaluate	Evaluate experiment
POST	/api/experiments/<name>/stop	Stop experiment
POST	/api/experiments/<name>/delete	Delete experiment