Dashboard Guide — Inference
How to use the Hypervize console for keys, elastic testing, and dedicated endpoint management.
Dashboard Guide — Inference
The Hypervize dashboard is the primary place to manage inference: API keys, the Elastic playground, and dedicated endpoints.
For API call patterns and endpoint behavior, see Dedicated Inference. For what happens when you Deploy, see Provisioning.
Accessing the Dashboard
After logging in you will land on the main resource overview. The left sidebar contains:
- Fleet Overview — High-level view of your inference blocks
- Inference — Elastic playground + dedicated deployment
- Keys — API key management
Billing, usage, and Alexandria (platform tools) live under Settings and related dashboard routes.
Inference Tab
This is the main screen for day-to-day inference work.
Elastic Playground
- Grouped model selector (by provider, color-coded)
- Chat interface with streaming
- Real-time telemetry (TTFB, prompt/completion tokens, estimated cost)
- Markdown rendering of responses
- Ability to switch between many models quickly
Use this to evaluate quality, latency, and pricing before committing to production traffic.
Dedicated Deployment Form
The right-hand side (or dedicated sub-tab) lets you:
- Choose source (Hugging Face is the primary path)
- Enter a model ID (e.g.
Qwen/Qwen2.5-7B-Instruct) - (Optional) Paste a Hugging Face token for gated models
- Name the endpoint
- Set min and max instance count (controls auto-scaling range; Min Nodes = 0 enables scale-to-zero)
- Optionally attach a custom domain (also available later via support)
- Toggle logging/telemetry
- Choose Public or Private auth mode
Live pricing estimate updates as you type the model ID (fetches HF metadata when possible).
After clicking Deploy you receive an endpt-... ID immediately. Capacity is provisioned asynchronously in the background; you also get email when the endpoint is ready or failed.
Inference Block Detail Pages
Click any of your dedicated endpoints from the overview or the list on the Inference page.
Each block shows:
- Current status with color coding (
ONLINE,PROVISIONING,SLEEPING,WAKING UP,FAILED, etc.). Scale-to-zero endpoints show SLEEPING when hibernated and WAKING UP when a request wakes them (status usually updates within about a minute during transitions). - Hourly burn rate (base + add-ons)
- Configuration summary
- Tabs: Overview, Logs, Playground
- Terminate — permanently tear down the endpoint (confirmation + progress overlay). Capacity is removed; charges already accrued that week still appear on the weekly invoice. See Dedicated Inference → Terminate.
The embedded playground on the detail page is pre-bound to that specific dedicated endpoint — useful for validation after provisioning and when an endpoint is waking from SLEEPING.
Keys Management
Located under Settings → Keys (also reachable from the sidebar).
- View all active keys (masked)
- Create new keys with name + scope
- Revoke keys instantly
Remember: every account starts with one default inference key.
Resource Overview (Fleet)
Shows aggregate burn rate, number of active inference blocks, and quick links focused on your inference endpoints.
Tips
- Use the playground heavily before sending production traffic.
- Keep the dedicated detail page open after deploying — watch status change from
PROVISIONINGtoONLINE. For scale-to-zero, you'll also see transitions to SLEEPING and WAKING UP (status usually updates within about a minute). - Create environment-specific keys (e.g., “prod-website”, “internal-agents”).
- Check the Logs tab first when debugging a dedicated endpoint.
- Use Terminate on the detail page when you are done with an endpoint (there is no in-place edit of model/min/max).