Dashboard

AgentLens includes a built-in web dashboard served by the same Hono process. Open http://localhost:3400 after starting the server.

Pages

Overview

The landing page shows a metrics grid with:

• Total sessions (with trend)
• Total events captured
• Active sessions count
• Error rate percentage
• Total estimated cost

Sessions

A filterable, sortable table of all agent sessions:

• Filters: Agent ID, status (active / completed / error), date range, tags
• Columns: Agent name, status, started at, duration, event count, tool calls, errors, cost
• Search: Full-text search across agent names and session IDs

Click any session to view its detail page.

Session Detail

A vertical timeline showing every event in chronological order:

• Each event shows: timestamp, event type icon, summary, duration
• Expand any event to see its full JSON payload with syntax highlighting
• Hash chain verification: A badge indicates whether the event chain is valid (tamper-evident)
• Color-coded by event type:
  • 🔵 Tool calls
  • 🟢 Tool responses
  • 🔴 Errors
  • 🟡 Approval events (from AgentGate)
  • 🟣 Form events (from FormBridge)
  • 💰 Cost tracking events

Events Explorer

A filterable list of all events across all sessions:

• Filters: Event type, severity, agent ID, session ID, date range
• Full-text search: Search within event payloads
• Detail panel: Click any event to see its full payload

Analytics

Charts and metrics for understanding agent behavior:

• Event trends: Events over time, grouped by hour/day/week
• Cost breakdown: Cost by agent, cost over time
• Agent comparison: Session count, error rate, avg duration per agent
• Tool usage: Most-used tools, error rates, average duration per tool

Alerts

Configure and manage alert rules:

• Create rules: Error rate exceeds threshold, cost spikes, latency anomalies, inactivity
• Alert history: View triggered and resolved alerts with timestamps
• Notification channels: Webhook URLs for external notifications

Health Overview

A dedicated view for agent reliability monitoring:

• Agent list — every agent with their overall health score (0–100), trend arrow, and session count
• Dimension breakdown — click any agent to see the 5-dimension radar: error rate, cost efficiency, tool success, latency, completion rate
• Historical chart — sparkline of daily health snapshots showing the trend over time
• Color coding — scores are green (≥75), yellow (50–74), or red (<50)

Cost Optimization

Actionable recommendations for reducing LLM costs:

• Recommendation cards — each shows current model → recommended model, complexity tier, projected monthly savings, and confidence level
• Summary metrics — total potential savings, analyzed calls, analysis period
• Success rate comparison — current vs. recommended success rates to assess risk
• Agent filter — narrow recommendations to a specific agent

Session Replay

Step-through debugger for past sessions:

• Timeline scrubber — drag or click to jump to any step in the session
• Step list — numbered, timestamped events with type icons (🤖 LLM, 🔧 tool, ⚠️ error)
• Context panel — at each step shows: cumulative cost, LLM conversation history, tool results, pending approvals, error count, warnings
• Event type filter — focus on specific interaction types
• Playback controls — play/pause, step forward/back, speed adjustment

Benchmarks

A/B experiment management:

• Benchmark list — all benchmarks with status badges (📝 draft, 🏃 running, ✅ completed, ❌ cancelled)
• Create form — define name, description, variants (name + tag), metrics, and optional time range
• Detail view — per-variant session counts, progress toward minimum sessions, lifecycle action buttons
• Results view — statistical comparison table with p-values, confidence stars, winner indication, and distribution charts per metric

Guardrails

Automated safety rules management:

• Rule list (/guardrails) — all guardrail rules with condition/action summaries, enabled toggle, trigger count, and last triggered time
• Create/Edit (/guardrails/new, /guardrails/:id/edit) — form for defining condition type, thresholds, action type, cooldown, dry-run mode, and agent scope
• Detail view (/guardrails/:id) — full rule configuration, runtime state (trigger count, last triggered, current value), and trigger history table
• Activity feed (/guardrails/activity) — real-time feed of all guardrail triggers across all rules with filtering by agent and rule name

Settings

Configure integration secrets and server settings:

• Retention policy (days)
• AgentGate webhook secret
• FormBridge webhook secret