Use Cases#
This page groups practical Admin UI scenarios and common use cases supported by OptScale AI. The examples below illustrate how the platform fits into different workflows and supports everyday operational and development tasks.
Dashboards#
Use Home → Dashboards for at-a-glance monitoring without opening Usage and Cost or Traces for every check. Start from the shared DEFAULT tab for organization-wide coverage, or + CREATE DASHBOARD for role-specific layouts. For panel types and the DEFAULT layout, see Core Services — Home / Dashboards and Dashboard panel types.
| Use case | Typical panels | Dashboard pattern |
|---|---|---|
| Daily platform health check | Total requests, Success rate, Avg latency | Open DEFAULT; set time range to 24h; use REFRESH or auto-refresh during active rollout |
| FinOps and spend review | Total spend, Cost breakdown, Models breakdown, Users breakdown | DEFAULT or a pinned custom tab; use 7d or 30d to review trends before budget meetings |
| Find top cost drivers | Models breakdown (Spend), Users breakdown (Spend) | Custom dashboard for finance or platform owners; rank by Spend to target model or user optimization |
| Track token growth | Total tokens, Total input tokens, Total output tokens | Compare cards and the Total tokens chart over 7d / 30d for capacity and quota planning |
| Governance and compliance monitoring | Top guardrail violations, Success rate | After policies and guardrails are enabled; use 7d to spot recurring violation types |
| Validate a new model or provider | Model reliability hotspots, Success rate, Avg latency | Custom or DEFAULT tab with 24h after rollout; drill into failures in Traces when hotspots appear |
| Incident or outage triage | Success rate, Total requests, Model reliability hotspots | Short window (1h); REFRESH frequently; pair with Traces for per-request detail |
| Evaluate prompt caching | Cache efficiency | Add to a custom FinOps dashboard when repeated prompts or shared context are in use |
| Executive or org-wide overview | DEFAULT summary cards and charts | PIN DASHBOARD on DEFAULT (or a curated custom tab) so operators land on the same view |
| Personal operator workspace | Mix of cards and charts for your responsibilities | + CREATE DASHBOARD on a MY tab; EDIT layout and Panel type per tile |
| ML training and experiment oversight | Recent tasks, Recent runs, Runs activity, Recent models, Model versions | Custom dashboard for ML leads; combine with Experiment Tracking for run detail |
| User adoption and activity | Users breakdown (Total tokens), Total requests | 7d / 30d on DEFAULT or a team-facing custom dashboard to spot heavy usage or idle accounts |
Allow time for new traffic to appear after you change providers, routing, or policies before expecting panels to update. Use Usage and Cost for tabular drill-down and Traces when a dashboard metric needs request-level evidence.
Policies + Guardrails#
| Use case | Typical guardrails | Policy pattern |
|---|---|---|
| Redact PII in user prompts | PII detection and redaction on Input, action Redact | Match Chat Completion (or your API request type); Stage Input; Sampling rate 100% for compliance-sensitive workloads |
| Block credential leaks | Secrets on Input and/or Output | Organization-wide or team-scoped conditions; block or redact before content reaches the provider or the client |
| Restrict unsafe or off-topic content | Ban topics, Toxicity, or Sentiment | Scope by Request type, Provider, or Team when only certain workloads need stricter rules |
| Harden against prompt abuse | Prompt injection, Jailbreak, Invisible text on Input | Apply to externally facing Chat or API traffic; start with moderate thresholds and tune from Violation rate |
| Control response safety | Toxicity, Code injection, or Secrets on Output | Stage Output so model responses are checked before they return to the user or calling application |
| Limit oversized requests | Token limit | Match high-volume or untrusted entry points; enforce limits before expensive provider calls |
| Layered governance | Multiple guardrails on one policy (for example PII detection and redaction + Secrets on Input) | One policy with shared Conditions; each linked guardrail keeps its own Type, Threshold, and action |
Reuse guardrails across policies when the same control applies to different scopes (for example one Secrets guardrail linked from both a Chat policy and an API policy with different conditions).