Dashboard Guide

Every agent, every request. Here's what each view shows and how to use it.

Overview

The landing page of the dashboard. Shows system-wide health at a glance:

• Total registered agents and their lifecycle stage distribution
• Actions in the last 24h / 7 days / 30 days
• Total cost in the same windows
• Recent activity feed: the last 20 run events across all agents
• Agents with the highest cost or error rate in the current period

Agents list

Every registered agent, sortable and filterable by:

• Name / Gateway key
• Connection type
• Owner (populated from verified OIDC when SSO is configured)
• Lifecycle stage (Development / Testing / Staging / Production / Degraded / Retired)
• Last active / last run
• Total cost (current billing period)

Agent detail

The per-agent view is split into four main tabs, plus a header containing the agent name and gateway key, owner, and a manual Revoke/Reinstate control:

1. Timeline

• Cost Summary: Total cost, tokens, runs, and average cost per run, plus a breakdown of cost by provider/model.
• Usage Trend: Visual charts of token usage and requests over time.
• Activity Timeline: A chronological event log showing every run (with input/output tokens and status).
• Authorization Panel: A ledger of the agent's access grants and scope changes.

2. Performance

Tracks four key operational pillars:

• Reliability: Success rate, error count, latency percentiles (P50/P95/P99), error type breakdown, and a chart of failure hotspots by action.
• Cost Efficiency: Total cost, cost per successful action, period-over-period cost trend, fleet percentile, and a donut chart of cost by model.
• Activity: Last active time, dormancy status (flagging inactive agents), and action volume trends.
• Security Hygiene: Token staleness, delegation fan-out/fan-in, and unused scope detection.

3. Governance & Scopes

• Visual management of what the agent is allowed to do.
• Toggle gateway scopes: the universal gateway:*, or granular gateway:<provider> and gateway:model:<model_name> (see Scopes).
• Assign custom string scopes for internal capabilities.

4. Budget & Limits

• Budget Cap (USD): Set a hard limit on daily or monthly spend. The gateway blocks new requests when exceeded.
• Rate Limit (RPM): Set a maximum number of requests per minute to prevent runaway loops.
• Live Monitoring: Progress bars showing current budget utilization and rate limit usage in real-time.

Cost & Usage

The cost page. All costs are stored values: computed once at write time, not recalculated on load. See Metrics Reference for the exact calculation.

• KPIs & Outliers: Total/Avg cost, plus P50 and P95 cost-per-run to detect outliers.
• Alerts & Savings: Visibility into Unpriced Actions and estimated Cache Savings.
• Trend Charts: Cost, Token Usage, and Runs over time. You can filter these charts to view all models or isolate a specific model.
• Cost by Provider & Model: A donut chart for providers and a detailed breakdown table for models.
• Cost by Agent Leaderboard: Drill down into which agents are spending the most.

Agent graph visualizer

Two layout modes, toggled with a button in the top-right corner:

• Grid layout (default): agents arranged in a responsive grid. Clean and readable for large fleets.
• Force-directed graph (d3-force): agents as nodes, with their relationships as edges. Useful for understanding how a fleet of agents fits together.

Lifecycle Kanban

A governance board for managing agent lifecycle stages. Six columns:

Drag agents between columns to update their stage. Stage transitions are logged with a timestamp and the identity of the person who moved them.

Per-agent timeline

A chronological event log for a single agent. Shows every request, access change, and lifecycle stage transition in time order. Useful for post-incident investigation: you can see exactly what happened, in what order, without writing any queries.

Next steps

• Metrics Reference: how cost and latency numbers are calculated
• SSO / OIDC: enable owner attribution