Welcome
Welcome is the operator landing page. It combines a greeting, a Lens Assistant prompt starter, current cluster issues and a small set of quick actions. It does not duplicate cluster capacity, model-service tables, deployment history or token usage; those remain on their owning pages.
Who sees Welcome
Welcome is the first sidebar item for admins, maintainers and custom roles with
cluster:cluster:read. Consumer-only roles keep their existing landing page.
The page always filters issues and actions to the clusters and permissions your
account can use. If no cluster is accessible and your role can create clusters,
Welcome offers to open the create-cluster wizard.
Join Slack
Click Join Slack in the Welcome header to open the llm-d Slack workspace in a new tab.
Open the documentation
Click Documentation in the Welcome header to open the Lens documentation in a new tab.
Star the project
Click Star on GitHub in the Welcome header. Lens opens GitHub sign-in and
returns to the llm-d-incubation/llm-d-lens repository after you sign in. Click
Star on GitHub to finish.
Ask Lens Assistant
- Type a message in Ask Lens Assistant.
- Choose an AI provider in the lower-left of the composer.
- Click Send, or press
Ctrl+Enter/Cmd+Enter.
Sending opens Lens Assistant, selects the provider you chose, and submits the message automatically once the provider is ready. Tool calls keep the assistant's normal confirmation rules. The prompt is carried in temporary browser session storage, not in the URL.
Only providers Lens Assistant can use are listed. If none are available, Send is disabled and the composer links to External providers. If your role cannot add providers, ask an administrator. If the provider list cannot be loaded, retry; do not assume no provider exists.
If your role cannot use Lens Assistant, the page shows a short permission note instead of an unusable composer. Issues and permitted quick actions remain available.
Review cluster issues
Needs attention lists current findings in the selected cluster scope:
- No accessible clusters, a cluster without model-cache storage, or a recorded HF_TOKEN Secret that no longer exists. The missing-storage issue links to New storage volume with the cluster and the Model cache purpose already selected.
- A model-cache volume has no cached model, or a cluster's model-cache volumes contain no ready model.
- Cluster components or nodes that are not ready. Missing Grafana dashboards are informational when core monitoring is healthy.
- Model services without a ready serving path, or unavailable Gateway/IPP.
- Sustained serving pressure or backend errors on a backing deployment.
- Recent deployment failures, or ready deployments that no model service publishes.
- Failed storage provisioning or model-cache download/sync.
- Active users with no direct or group-inherited role, or groups with no role binding (info). Each user finding names the user and links to that user's role editor.
- No compatible External AI provider (info).
Each row shows severity, resource identity, evidence, the first time the issue was seen and a diagnostic link labeled To check, To fix, or To resolve. A finding without sufficient telemetry is shown as unknown or stale, not as healthy.
Use the cluster selector in the section header to narrow the list. View all issues opens the full, filterable issue page. The page refreshes current issues every 15 seconds and keeps the existing rows during a quiet refresh.
Quick actions
The right-hand rail contains actions your role can perform:
| Action | What it opens |
|---|---|
| Token usage analytics | Usage reports for recorded token consumption. |
| Add a model deployment | Model market, to prepare a model for serving. |
| Publish a model service | Model services with the add form requested. |
| Launch benchmark | Evaluation with the new-task wizard open. |
| Replay traces | Simulation with the new-task form open. |
| Manage storages | Storage, to inspect or create cluster volumes. |
| Cache a new model | Model cache with the download form requested. |
| Create a new cluster | Clusters with the create wizard requested. |
Actions are navigation only. Forms, validation, confirmations and writes remain on the destination pages and are still permission-gated. Actions you cannot perform are not shown.
Empty and partial states
| What you see | Meaning |
|---|---|
| All systems operational | Health checks and telemetry across monitored clusters are reporting normally. No action is required. |
| Cluster access needed | Your account has no accessible cluster. Ask for a cluster-scoped role. |
| Some checks unavailable | One or more telemetry or Kubernetes sources did not answer. |
| No compatible provider | Add an OpenAI-compatible provider under External providers. |
| Assistant not available | Your role lacks playground:chat:use. |