Skip to main content
The dashboard at app.orbitrage.ai organizes everything around three levels — Organization, Workflow, and End-user — so you can zoom from “total spend this month” down to “what one customer’s agent did on one run.”

​
Overview

Your org-wide home page: headline KPIs (calls, cost, savings, incidents), cost-trend with anomaly markers, and per-project health. Start here to see whether anything needs attention.
Orbitrage Overview dashboard

​
Workflows

A workflow is a project — every API key belongs to one. The list shows runs/month, incidents, cost, and status (running / paused / degraded) per project, and lets you create new ones. Open a workflow to get its flow graph: an interactive reconstruction of each run, with a node per LLM call and tool invocation. Click a node for messages, model, tokens, cost, latency, routing signals, and fallback chain. The detail page also has tabs for Analytics, Incidents, Intelligence, and Settings.
Workflow flow graph with model, subagent, and tool-call nodes
The Analytics tab breaks down calls, unique users, cost by token type, top models, and (inferred) user sentiment — with an Ask-Analytics assistant for ad-hoc questions:
Project analytics with Ask Analytics assistant

​
Routing

Every router decision, live. A sortable table of calls — requested model, routed model, tier, tokens, latency, cost, savings — plus charts for cost/latency per call, model distribution, and tier breakdown. Click any row for the full routing detail, including the score and the signals that drove it.

​
Telemetry & Traces

  • Telemetry — the live span stream from your calls, with per-span detail (trace id, provider, tokens/sec, baseline cost) and org-wide totals including “what you’d save with the router.”
  • Traces — an org-wide drilldown across all workflows by Requests, Errors, Models, and Tools.

​
The three insight clusters

Within Overview and each workflow, analytics group into:

Cost

Daily spend with anomaly detection, trend forecast, cost composition, a money-flow view, an hour-of-day heatmap, and top models / providers / users.

Topology

The routing matrix, context-window distribution, run shapes, a tool leaderboard with co-occurrence, and end-user similarity.

Errors

Per-provider error rates with confidence intervals, daily error series, refusal & truncation rates, and finish-reason drift.

​
Models

Browse the routable catalog with per-model pricing and tier, test any model in a chat modal, and manage BYOK provider keys — including a 30-day BYOK spend tracker and an org-level “fall back to pooled on BYOK failure” toggle.

​
Playground

A prompt-iteration lab: build multi-step chains, A/B test branches, run them against real traffic samples, diff versions, and promote a winner. See Playground.

​
Keys, Billing, Settings

  • Keys — mint, list, and revoke API keys (each scoped to a workflow).
  • Billing — credit balance, usage, and top-ups. See Billing.
  • Customise — manage webhook alerts, pinned charts, MCP keys, and the Slack connection.
  • Settings / Workspace / Orgs — profile, organization settings, and switching between organizations.
Prefer chat? Almost everything here — analytics, projects, keys, alerts — can also be driven from Slack or Ask Analytics without opening the dashboard.