ATLAS
Admin AI Usage & Cost Dashboard

Monitor all AI integrations: Gemini Flash (text), Gemini Pro Image (design), GPT Image 1.5 (avatars). Cost tracking, rate limits, error monitoring, and optimization.

1 Overview
2 Model Detail
3 Cost & Optimization
Overview
AI Usage Overview
9:41
AI Usage Live
Today
7 Days
30 Days
Custom
2,847
API Calls Today
+12.4%
$342.18
Cost MTD
-3.2%
$0.28
Cost Per User
+1.1%
99.2%
Success Rate
+0.3%
Model Breakdown
3
Models
Gemini Flash — 60%
Gemini Pro Image — 22%
GPT Image 1.5 — 18%
ModelsView All
Gemini 3 Flash
1,708 calls • $48.20
Healthy
Gemini Pro Image
626 calls • $187.80
Healthy
GPT Image 1.5
513 calls • $106.18
Elevated
Requests Today24h
12 AM 6 AM 12 PM 6 PM Now
AIUsageOverview screen — useAIUsageStats(period) hook with real-time subscription. Donut via ModelBreakdownChart.
Model Detail
Gemini 3 Flash
9:41
Gemini 3 Flash Healthy
gemini-3-flash-preview
Text & Chat • Google AI
1,708
Requests/Day
+8.2%
$48.20
Cost Today
-2.1%
142ms
Avg Latency
-5ms
0.3%
Error Rate
-0.1%
Token UsageToday
1.2M
Input Tokens
480K
Output Tokens
71% input 29% output
Requests / Day7 days
Mon
1,412
Tue
1,564
Wed
1,738
Thu
1,477
Fri
1,629
Sat
912
Sun
1,708
Top Use Cases
AI Assistant
892 calls • Avg 320 tokens
52%
Job Detection
478 calls • Avg 180 tokens
28%
Sentiment Analysis
338 calls • Avg 95 tokens
20%
ModelDetailScreen — useModelMetrics("gemini-3-flash"). Token bar via TokenUsageBar. Chart via RequestsBarChart.
Gemini Pro Image
9:41
Gemini Pro Image Healthy
gemini-3-pro-image-preview
Design Studio • Google AI
626
Generations/Day
+15.7%
$0.30
Cost / Generation
Same
94.8%
Success Rate
+1.2%
8.4s
Avg Gen Time
-0.6s
Generations / Day7 days
Mon
480
Tue
541
Wed
610
Thu
506
Fri
567
Sat
349
Sun
626
Top Styles Requested
Modern Interior
187 generations • 8.1s avg
30%
Exterior Renovation
144 generations • 9.2s avg
23%
Kitchen Remodel
119 generations • 8.8s avg
19%
Recent GenerationsGallery
ModelDetailScreen — useModelMetrics("gemini-3-pro-image"). Gallery via RecentGenerationsGrid with lazy image loading.
GPT Image 1.5
9:41
GPT Image 1.5 Elevated
gpt-image-1.5
Avatar Generation • OpenAI
Queue depth elevated
23 avatars queued — avg wait 45s
513
Avatars/Day
+22.1%
$0.21
Cost / Avatar
-$0.02
23
Queue Depth
+8
4.2%
Retry Rate
+1.1%
Resolution Breakdown
512x512
308
1024x1024
154
256x256
51
Queue StatusLive
Processing
5 avatars
Active
Waiting
18 avatars
Queued
Retries Today
22 attempts
4.2%
Completed Today
491 avatars
95.8%
ModelDetailScreen — useModelMetrics("gpt-image-1.5"). Queue via AvatarQueueStatus with real-time sub.
Cost & Optimization
Cost Breakdown
9:41
Cost Breakdown
7 Days
30 Days
90 Days
YTD
78% of monthly budget used
$342.18 of $440 budget — 6 days remaining
Budget Progress
$0 $342 / $440 $440
Cost by Model30 days
Wk1
$78
Wk2
$84
Wk3
$92
Wk4
$88
Flash
Pro Image
GPT Image
Projected Monthly
$427.73
Projected at current rate
Near Budget
Cost by Feature
$342
MTD
Design Studio — 42%
AI Assistant — 28%
Avatars — 20%
Detection — 10%
Per-Model Costs
Gemini Flash
$124.60
36.4%
Gemini Pro Image
$143.70
42.0%
GPT Image 1.5
$73.88
21.6%
CostBreakdownScreen — useAICostMetrics(period). Stacked chart via CostStackedBar. Budget line via BudgetProgressBar.
Rate Limits & Quotas
9:41
Rate Limits Live
Current Usage vs Limits
Gemini Flash
1,708 / 10,000 RPD • 28 / 60 RPM
17%
Gemini Pro Image
626 / 1,000 RPD • 12 / 15 RPM
63%
GPT Image 1.5
513 / 1,000 RPD • 8 / 10 RPM
51%
Burst Capacity
72%
Flash Burst
28%
Pro Image Burst
38%
GPT Burst
Auto-Scaling
Gemini Flash Scaling
Auto-increase RPM on demand spikes
Image Gen Queuing
Queue overflow requests instead of rejecting
Fallback Models
Switch to backup model on rate limit
Recent Throttle EventsView All
GPT Image 1.5 — RPM limit hit
2 min ago • 3 requests queued
Throttled
Gemini Pro Image — RPM limit hit
18 min ago • 1 request queued
Resolved
Flash auto-scaled — 60 to 90 RPM
1 hr ago • Scaled back after 12 min
Resolved
RateLimitsScreen — useQuotaMetrics() real-time. Quota bars via QuotaGaugeRow. Throttle log via ThrottleEventFeed.
Error Monitoring
9:41
Error Monitoring
1.8%
Error Rate (24h)
+0.3%
87%
Retry Success
+2.1%
Error Rate Timeline24h
12 AM 6 AM 12 PM 6 PM Now
Error Types24h
Timeout
18
Rate Limit
14
Content
8
Bad Input
5
Recent ErrorsView All
Timeout — GPT Image 1.5
Request exceeded 30s timeout during avatar generation for user usr_2kJ9xM
at generateAvatar() → openai.images.generate() → timeout after 30012ms
2 min ago • Retried OK
Content Filter — Gemini Pro Image
Generation blocked by safety filter. Prompt contained flagged terms for design request #DR-4821
at generateDesign() → gemini.generateContent() → SAFETY_BLOCK
8 min ago • User notified
Rate Limit — Gemini Pro Image
429 Too Many Requests. RPM quota exhausted at 15/15. Queued 3 pending requests.
at designStudio() → gemini.generateImages() → 429 RESOURCE_EXHAUSTED
18 min ago • Auto-resolved
Invalid Input — Gemini Flash
Token limit exceeded (32,768 max). User sent 41,200 tokens in AI assistant conversation.
at aiAssistant() → gemini.chat() → INVALID_ARGUMENT: max_tokens
34 min ago • Truncated & retried
Retry Performance
87%
Retry Success
1.3
Avg Retries
2.4s
Avg Wait
ErrorMonitoringScreen — useAIErrorMetrics() real-time subscription. Error log via ErrorLogFeed. Timeline via ErrorRateTimeline.
Optimization
9:41
Optimization
Estimated Monthly Savings
$68.40
4 recommendations • 16% cost reduction
Cache Performance
Response Cache
Identical prompts served from cache
34%
Hit Rate
952
Cache Hits
$18.20
Saved MTD
Recommendations
Optimize Prompt Length
-$24.50/mo
AI Assistant prompts average 1,200 tokens. Trimming system prompts and using structured few-shot examples could reduce input tokens by 35% with no quality loss.
High Impact Low Effort
Downgrade Avatar Resolution
-$18.90/mo
60% of avatars render at 512x512 but are displayed at 128x128 in the app. Generating at 256x256 would cut GPT Image costs by 35% with no visible quality difference.
High Impact Low Effort
Batch Design Requests
-$15.00/mo
Design Studio makes 3 separate API calls per generation (style, render, upscale). Batching into a single multi-step call reduces overhead and latency by 40%.
Medium Impact Medium Effort
Increase Cache TTL
-$10.00/mo
Current cache TTL is 1 hour. Extending to 24 hours for sentiment analysis and job detection would increase hit rate from 34% to an estimated 52%, saving ~$10/mo in redundant calls.
Medium Impact Low Effort
Savings Breakdown
Prompts
$24.50
Avatars
$18.90
Batching
$15.00
Cache
$10.00
OptimizationScreen — useAIOptimizations() computes savings from usage patterns. Recs via RecommendationCard. Cache stats via CachePerformanceCard.

Data Architecture

Backend schemas, queries, hooks, and components powering the AI Usage Dashboard.

aiUsageLogs (Schema)

_id: Id<"aiUsageLogs">
timestamp: number // Date.now()
model: "gemini-3-flash" | "gemini-3-pro-image" | "gpt-image-1.5"
feature: "ai_assistant" | "job_detection" | "sentiment" | "design_studio" | "avatar"
status: "success" | "error" | "cached" | "retried"
inputTokens: v.optional(v.number())
outputTokens: v.optional(v.number())
cost: number // USD cents
latencyMs: number
userId: v.optional(v.string())
errorType: v.optional("timeout" | "rate_limit" | "content_filter" | "invalid_input")
errorMessage: v.optional(v.string())
retryCount: v.optional(v.number())
cached: boolean // default false
resolution: v.optional(v.string()) // "256x256" | "512x512" | "1024x1024"
metadata: v.optional(v.any())

aiQuotas (Schema)

_id: Id<"aiQuotas">
model: "gemini-3-flash" | "gemini-3-pro-image" | "gpt-image-1.5"
rpmLimit: number // requests per minute
rpdLimit: number // requests per day
currentRpm: number
currentRpd: number
burstCapacity: number // percentage remaining
autoScale: boolean
fallbackEnabled: boolean
lastUpdated: number

aiBudgets (Schema)

_id: Id<"aiBudgets">
month: string // "2026-02"
budgetCents: number // $440 = 44000
spentCents: number
alertThreshold: number // 0.75 = 75%
alertSent: boolean
projectedCents: number

// Per-model breakdown
flashSpent: number
proImageSpent: number
gptImageSpent: number

aiOptimizations (Schema)

_id: Id<"aiOptimizations">
type: "prompt_trim" | "resolution_downgrade" | "batch" | "cache_ttl"
title: string
description: string
estimatedSavingsCents: number
impact: "high" | "medium" | "low"
effort: "low" | "medium" | "high"
status: "pending" | "applied" | "dismissed"
appliedAt: v.optional(v.number())
appliedBy: v.optional(v.string())

Convex Queries & Mutations

// api.admin.aiUsage
overview({ period: "today" | "7d" | "30d" | "custom" })
modelDetail({ model: string, period })
costBreakdown({ period })
errorLog({ model?, errorType?, limit })
quotaStatus() // real-time quota snapshot
optimizations() // computed recommendations

// api.admin.aiUsage (mutations)
applyOptimization({ optimizationId })
dismissOptimization({ optimizationId })
updateBudget({ month, budgetCents, alertThreshold })
toggleAutoScale({ model, enabled })
toggleFallback({ model, enabled })

Hooks & Components

// Hooks
useAIUsageStats(period) // overview KPIs + model breakdown
useModelMetrics(model) // per-model detail
useAICostMetrics(period) // cost breakdown + budget
useQuotaMetrics() // real-time quota gauges
useAIErrorMetrics() // error feed + timeline
useAIOptimizations() // computed savings recs

// Screen Components
AIUsageOverview // KPIs + donut + model cards
ModelDetailScreen // per-model deep dive
CostBreakdownScreen // stacked bars + budget + pie
RateLimitsScreen // quota bars + throttle log
ErrorMonitoringScreen // timeline + error feed
OptimizationScreen // recs + cache stats

// Shared Components
ModelBreakdownChart // donut/conic-gradient
TokenUsageBar // input/output split bar
RequestsBarChart // horizontal bar chart
CostStackedBar // stacked segments by model
BudgetProgressBar // fill + threshold marker
QuotaGaugeRow // usage vs limit bar
ThrottleEventFeed // throttle log list
ErrorRateTimeline // color-coded bar timeline
ErrorLogFeed // error list + stack traces
RecommendationCard // optimization rec with actions
CachePerformanceCard // hit rate + saved $
RecentGenerationsGrid // lazy image grid
AvatarQueueStatus // live queue depth
1 Usage Overview + Model Stats
2 Cost & Rate Limits
3 Errors & Optimization
Overview & Model Stats
Usage Overview + Model Stats
9:41

AI Usage Overview & Model Performance

AI Usage Overview
Total API Calls
Today 12,847
This Week 89,432
This Month 342,156
Avg Latency 245ms
Model Distribution
Gemini Flash 68%
Gemini Pro Image 24%
GPT Image 1.5 8%
Cost Summary
Today $18.42
MTD $487.23
Budget Used 64%
Model Performance
Gemini 3 Flash
Requests 8,741
Avg Latency 142ms
Error Rate 0.3%
Cost/1K $0.08
Gemini Pro Image
Requests 3,284
Avg Latency 1.2s
Error Rate 1.1%
Cost/1K $2.40
GPT Image 1.5
Requests 822
Avg Latency 3.8s
Error Rate 0.8%
Cost/1K $4.80
Cost Analysis & Rate Limits
Cost & Rate Limits
9:41

Cost Breakdown & Rate Limits & Quotas

Cost Breakdown
Monthly Spend
Gemini Flash $127.40
Gemini Pro Image $284.60
GPT Image 1.5 $75.23
Total MTD $487.23
Budget Status
Monthly Budget $750.00
Remaining $262.77
Projected $689.42
Status On Track
Rate Limits & Quotas
Gemini Flash Limits
RPM 1,200 / 1,500
TPM 892K / 1M
Daily 45K / 50K
Status Healthy
Image Gen Limits
RPM 42 / 60
Daily 2,100 / 3,000
Queue Depth 3
Status Normal
Error Monitoring & Optimization
Errors & Optimization
9:41

Error Monitoring & Optimization

Error Monitoring
Error Summary
Total Errors (24h) 47
Rate Limit 23
Timeout 12
Auth Failure 8
Other 4
Recent Errors
10:42 AM Rate limit — Flash
10:38 AM Timeout — Pro Image
10:15 AM 429 — Flash burst
Optimization
Cache Performance
Hit Rate 73%
Saved Calls 18,420
Cost Saved $142.30
Avg TTL 4h
Recommendations
Enable response caching for Flash Save ~$45/mo
Batch avatar generation queue Reduce timeouts 40%
Upgrade to Flash 2 tier Lower cost/req
1 AI Operations Dashboard
2 Model Detail & Analytics
AI Operations Dashboard
AI Operations Dashboard
https://admin.3bids.io/ai-usage
3Bids Admin
admin.3bids.io
MAIN
Dashboard
Models
Costs
Rate Limits
Errors
Optimization
Settings
DA
Danny Admin
Platform Admin

AI Usage Dashboard

3 Models Active Budget On Track
Overview
Dashboard
Complete ai usage dashboard with real-time data, analytics, and management tools.
Quick Actions
Full desktop layout combines all mobile screens into a dense, actionable dashboard with sidebar navigation, data tables, charts, and management controls.
Web DashboardThis web view consolidates all mobile screens into a single responsive dashboard. Sidebar navigation replaces tab-based mobile flow, with data density optimized for desktop viewport.
Model Detail & Analytics
Model Detail & Analytics
https://admin.3bids.io/ai-usage/models
3Bids Admin
admin.3bids.io
MAIN
Dashboard
Models
Costs
Rate Limits
Errors
Optimization
Settings
DA
Danny Admin
Platform Admin

Model Analytics

Gemini Flash 1 Alert
Overview
Dashboard
Complete model analytics with real-time data, analytics, and management tools.
Quick Actions
Full desktop layout combines all mobile screens into a dense, actionable dashboard with sidebar navigation, data tables, charts, and management controls.
Web DashboardThis web view consolidates all mobile screens into a single responsive dashboard. Sidebar navigation replaces tab-based mobile flow, with data density optimized for desktop viewport.