The web console

What each screen is for, and which ones you will actually use.

The console is where you watch work, compare results and govern the platform. Submitting still happens from sf — both talk to the same API, so nothing is console-only by accident.

Dashboard: cluster GPUs by series, fleet GPU-hours, and the submission heatmap.

Every screen

Screen	Path	What it is for
Dashboard	/	Cluster watermark, your quota, active jobs, recent submissions
Jobs	/jobs	The job list, with filters and bulk actions
Job detail	/jobs/:id	Charts, logs, validation, system, config, artifacts, diagnosis
Review	/review	Overlay several runs and decide which one to keep
Runs	/runs	The submission ledger, time-ordered
Projects	/projects	Runs grouped by repository, with project memory
Benchmarks	/benchmarks	The score matrix across runs and benchmarks
Rubrics	/rubrics	Written scoring standards, their versions and usage
Environments	/environments	Agent RL environments and what each declares
Playground	/playground	Chat with a checkpoint, and A/B two of them
Models	/models	The registry: versions, promotion, acceptance cards
Deployments	/deployments	Managed serving, revisions, tokens, Reflow
Methods	/methods	The method catalog and every tunable hyperparameter
Data	/data	Datasets and volumes: versions, file manifests, data quality
Plugins	/plugins	The extension shelf, and admin governance
Usage	/usage	Your quota, GPU-hours, and cost attribution
MCP	/mcp	Connect Cursor or Claude to your job data
Integrations	/integrations/hf	Link a Hugging Face account
Profile	/profile	Display name, password, signed-in devices
Notifications	bell icon	Your message feed
Sharing	—	Read-only public links to a job or a review
Tutorial	/tutorial	The in-product onboarding walkthrough

Administrators additionally get these, behind the Administration entry at the bottom of the sidebar — opening it replaces the sidebar with the admin navigation:

Screen	Path	What it is for
Users and quotas	/admin/users	Accounts, roles, GPU and storage quotas
Settings	/admin/settings	Platform switches, images, hardware registry, schedule windows
Scheduled tasks	/admin/tasks	Background worker schedules and run history
Audit	/admin/audit	Who submitted, who was denied, who changed what

Two things worth knowing early

Search is everywhere

⌘K opens a search palette over jobs, runs, datasets and models. It is usually faster than navigating.

Locked cards are features, not bugs

AI diagnosis, Ask Agent and MCP show a locked card when the deployment restricts them to administrators. FORGE_PREMIUM_PUBLIC opens them to everyone.