HiveBoard — Insights Guide #

Version: 0.2.0 Last updated: 2026-02-15

38 questions, four narrative moments — what your fleet data tells you that you should know.

Table of Contents #

1. Overview
2. Toolbar and Controls
3. Moment 1: The Glance
4. Moment 2: The Investigation
5. Moment 3: The Optimization
6. Moment 4: The Review
7. Question Reference
8. Data Sources

1. Overview #

The Insights page (insights.html) is an intelligence briefing for your AI agent fleet. Rather than showing raw data or configurable dashboards, it organizes 38 computed questions into four narrative "moments" — each designed for a specific mindset.

The page is designed to be read top-to-bottom like a report: start with a glance at fleet health, investigate specific agents, find optimization opportunities, and end with a holistic review.

Each moment is a collapsible section. Each question within a moment is rendered as a card or visualization with a question ID (Q1–Q38) shown in the corner for reference.

Key distinction from Analytics: Analytics lets you explore data freely across six analytical dimensions. Insights tells you what the data means — it's opinionated, narrative-driven, and includes automated detectors and health scoring.

2. Toolbar and Controls #

Unlike Analytics (which auto-refreshes every 60s), Insights refreshes on demand via the Refresh button. This is intentional — Insights is designed for deliberate review, not continuous monitoring.

3. Moment 1: The Glance #

Section title: "The Glance — Fleet Status at a Glance"

This moment answers: "Is everything OK? Do I need to act?" It's designed to be scannable in under 5 seconds.

Q1: Fleet Status Bar #

A horizontal bar segmented by agent status (processing, idle, stuck, error, waiting). Each segment shows the count and is proportionally sized. Below the bar, a legend labels each status with its count.

Read it as: Green/blue segments = healthy. Any red segments = action needed.

Q2: Fleet Vitals #

Four big-number cards:

Q3: Cost Snapshot #

Three cards for cost awareness:

Q4: Throughput Sparkline #

A mini bar chart showing task completion volume over time. Helps spot drops or spikes in activity.

Q5: Latest Activity Feed #

The 10 most recent events across the fleet, showing timestamp, agent name, event type, and summary. Includes a live badge if events are very recent.

4. Moment 2: The Investigation #

Section title: "The Investigation — Agent & Task Deep Dive"

This moment answers: "Let me dig into a specific agent — what's its story?"

Agent Selector #

A prominent selector bar at the top of this section lets you choose which agent to investigate. Styled with the accent color to distinguish it from the global agent filter.

Q6: Agent Status Card #

A detailed card for the selected agent showing:

• Status badge (Processing/Idle/Stuck/Error)
• Heartbeat age with colored indicator
• Current task ID (if processing)
• Last event type and time

Q7: Agent Performance Summary #

Key performance numbers for the selected agent in the current range:

• Tasks completed / failed
• Success rate
• Average duration
• Total LLM cost
• LLM calls per task average

Q8: Agent vs Fleet Comparison #

Side-by-side comparison of the selected agent against fleet averages:

Values better than fleet average are highlighted green; worse values are highlighted red.

Q9: Agent Task Breakdown #

Distribution of the selected agent's tasks by type, shown as horizontal bars with counts and percentages.

Q10–Q16: Extended Agent Analysis #

Additional cards covering:

• Q10: Error breakdown for this agent (by type)
• Q11: LLM model usage distribution
• Q12: Top actions/tools by invocation count
• Q13: Queue depth and pipeline status
• Q14: Cost trend over the selected range
• Q15: Duration trend over the selected range
• Q16: Task completion timeline

5. Moment 3: The Optimization #

Section title: "The Optimization — Cost & Reliability"

This moment answers: "Where can I improve? What should I fix or optimize?"

Q17: Cost Leaderboard #

Ranking of agents by total cost, with the most expensive highlighted. Shows cost, call count, and share of fleet total.

Q18: Cost per Task Ranking #

Agents ranked by average cost per task. Useful for finding agents that are expensive per unit of work (even if their total spend is modest due to low volume).

Q19: Model Cost Comparison #

Breakdown of cost by LLM model across the entire fleet. Helps identify whether switching models could save money.

Q20: Token Efficiency #

Analysis of input/output token ratios. Agents with very high input-to-output ratios may be sending oversized prompts.

Q21: Slowest Tasks #

A table of the slowest tasks in the fleet, showing task ID, agent, duration, and status. Helps identify performance bottlenecks.

Q22: Error Hotspots #

Agents ranked by error count with error rate percentages. Includes breakdown by error type.

Q23: Retry Analysis #

If agents use retries, shows retry frequency and which tasks/actions trigger the most retries.

Q24: Action Performance Ranking #

Actions ranked by success rate and average duration. Low success rates and high durations are highlighted as optimization opportunities.

Q25: Idle Time Analysis #

How much time agents spend idle vs. working. High idle percentages may indicate over-provisioning.

Q26: Queue Health #

Analysis of queue depths and item ages across agents. Long queues or stale items indicate capacity problems.

Smart Detectors (Q27–Q28) #

Automated checks that flag specific operational problems:

Each detector shows as a card with a colored left border: green (OK), amber (Warning), red (Error). The card includes a description of what was checked and the current finding.

6. Moment 4: The Review #

Section title: "The Review — Trends & Fleet Health"

This moment answers: "Stepping back — how is the fleet doing overall, and is it ready to scale?"

Q29–Q35: Metric Summaries #

Quick-reference cards for fleet-wide totals:

Q36: Trend Summary #

Computes directional trends (up/down/flat) for four key metrics over the selected time range:

Each trend shows a percentage change and a colored indicator. A verdict banner at the bottom summarizes:

Requires at least 4 timeseries data points to compute. Shows "Not enough data points" otherwise.

Q37: Health Gauge #

A visual gauge (0-100) representing overall fleet health. The score starts at 100 and subtracts for problems:

The gauge displays as a semi-circular SVG with color coding:

Below the gauge, a deduction breakdown shows exactly what reduced the score (e.g., "1 stuck (-15) · Success rate 82% (-4)"). If no deductions: "No deductions — everything looks great!"

Q38: Scale Readiness Checklist #

A pass/fail checklist evaluating whether the fleet is ready to scale:

Summary banner:

7. Question Reference #

Quick lookup of all 38 questions by ID:

8. Data Sources #

The Insights page pulls from seven API endpoints:

All data is fetched on page load and when the Refresh button is clicked. The range selector controls the time window for all time-scoped queries.