Usage Analytics

Track and analyze your Aize Platform usage with comprehensive metrics and insights.

Overview

The Usage & Analytics page provides detailed insights into:

• API request volumes
• Token consumption
• Cost breakdown
• Model usage patterns
• Performance metrics
• Team activity (for Admins/Owners)

Key Metrics

Total Requests

What it shows:

• Lifetime count of all API calls
• Includes successful and failed requests
• Updates in real-time

Useful for:

• Understanding overall usage volume
• Capacity planning
• Rate limit monitoring
• Trend analysis

View breakdown by:

• Time period (daily, weekly, monthly)
• Model
• API key
• Team member (Admins/Owners)
• Status code (success/error)

Total Tokens

What it shows:

• Combined input + output tokens
• Lifetime aggregation
• Separate input/output breakdown available

Token types:

• Input tokens: Text you send in prompts
• Output tokens: Generated responses
• Total tokens: Sum of both

Why it matters:

• Output tokens typically cost 2-3x input tokens
• Helps optimize prompt engineering
• Indicates data volume processed

Conversion guide:

• ~1 token = 4 characters
• ~1 token = 0.75 words
• ~100 tokens = 75 words
• ~1000 tokens = ~750 words or 1-2 pages

Total Cost

What it shows:

• Lifetime spending in USD
• Sum of all usage deductions
• Excludes deposits and refunds

Cost components:

Request Cost = (Base Model Cost × Tokens) × Markup + Fixed Fee

View breakdown:

• Cost per model
• Cost per API key
• Cost per team member
• Cost trends over time

Current Spend

What it shows:

• Current billing period spending
• Resets based on organization tier
• Progress towards any budget limits set

Color coding:

• Green: Under 75% of budget
• Yellow: 75-90% of budget
• Red: Over 90% of budget

Request Logs

Detailed view of every API call made to the gateway.

Log Record Information

Each request record includes:

Request Identification:

• Request ID: Unique UUID for tracing
• Timestamp: Exact date and time
• API Key: Which key was used (label shown)
• User: Member who made request (team keys only)

Model Information:

• Model Alias: What you requested (e.g., "gpt-4")
• Provider Model: Actual model used (e.g., "gpt-4-0613")
• Endpoint: API endpoint called (/v1/chat/completions)

Token Usage:

• Input Tokens: Tokens in your prompt
• Output Tokens: Tokens in response
• Total Tokens: Combined count

Cost Breakdown:

• Provider Cost: What we pay the AI provider
• Customer Cost: What you're charged
• Markup: Our markup (transparent)
• Fixed Fee: Per-request fee (if applicable)

Performance:

• Latency: Time to complete request (milliseconds)
• Status Code: HTTP response code (200, 401, 402, etc.)
• Error Message: Details if request failed

Metadata (expandable):

• Model parameters (temperature, max_tokens, etc.)
• Full request/response (if logging enabled)
• Provider-specific details

Filtering Logs

By Time Period:

• Last hour
• Last 24 hours
• Last 7 days
• Last 30 days
• Custom date range

By Status:

• All requests
• Successful only (2xx)
• Client errors (4xx)
• Server errors (5xx)

By Model:

• Filter to specific model
• Compare across models
• Identify most-used models

By API Key:

• View logs for specific key
• Useful for debugging
• Track key usage patterns

By User (Admins/Owners only):

• See team member activity
• Identify high-usage users
• Track individual costs

"My Logs Only" (Members):

• Toggle to see only your requests
• Hides team activity
• Privacy-focused view

Pagination

• Page size: 50 records per page
• Load more: Click to see older records
• Export: Download as CSV (coming soon)

Cost Analysis

Understanding Costs

Provider Cost vs Customer Cost:

Provider Cost: $0.10      (what we pay)
Markup (25%):   $0.025    (our margin)
Fixed Fee:      $0.01     (per-request fee)
─────────────────────────
Customer Cost:  $0.135    (what you pay)

Why the difference?:

• Infrastructure costs
• Support and maintenance
• Payment processing fees
• Service reliability guarantees
• Additional features (caching, routing, fallbacks)

Transparency:

• Both costs shown in request logs
• Clear markup percentage
• No hidden fees
• Predictable pricing

Cost Optimization Tips

1. Choose the Right Model:

• Simple tasks: Use GPT-3.5 Turbo (~10x cheaper than GPT-4)
• Complex reasoning: Use GPT-4 or Claude
• Code generation: Consider GPT-4 Turbo
• General chat: GPT-3.5 Turbo sufficient

2. Optimize Prompts:

• Be concise but clear
• Remove unnecessary examples
• Use system messages efficiently
• Avoid redundant instructions

3. Control Output Length:

• Set appropriate max_tokens
• Don't request more than needed
• Output tokens cost more than input

4. Use Streaming:

• Same cost, better UX
• Users see results faster
• Can cancel if sufficient

5. Implement Caching:

• Cache common responses
• Reduce repeat requests
• Significant savings for high-traffic apps

6. Set Budget Limits:

• Per-key monthly limits
• Team member budgets
• Alert before overspending

Cost Per Request Analysis

Average cost:

Average = Total Cost / Total Requests

Track trends:

• Is average increasing? (using more expensive models?)
• Sudden spikes? (longer outputs or errors?)
• Compare against budget expectations

Example averages by model:

• GPT-3.5 Turbo: $0.001 - $0.005 per request
• GPT-4: $0.01 - $0.05 per request
• Claude 3 Opus: $0.015 - $0.075 per request

Performance Analytics

Latency Metrics

What we measure:

• Time from request received to response complete
• Includes:
  • API processing time
  • Model inference time
  • Network latency
  • Queue wait time (if any)

Typical latencies:

• Fast models (GPT-3.5): 500-2000ms
• Standard models (GPT-4): 2000-5000ms
• Large outputs: Can be 10+ seconds

Monitoring:

• Average latency per model
• P50, P95, P99 percentiles (coming soon)
• Identify slow requests
• Track performance trends

Success Rate

HTTP Status Codes:

• 2xx Success: Request completed successfully
• 4xx Client Error: Problem with your request
• 5xx Server Error: Issue on our side or provider

Track success rate:

Success Rate = (2xx responses / Total responses) × 100%

Target: > 99% success rate

Common errors:

• 401 Unauthorized: Invalid API key
• 402 Payment Required: Insufficient balance
• 403 Forbidden: Model not allowed for key
• 429 Too Many Requests: Rate limit exceeded
• 500 Internal Server Error: Temporary issue
• 503 Service Unavailable: Provider down

Error Analysis

View errors:

1. Go to Request Logs
2. Filter by status code (4xx, 5xx)
3. Review error messages
4. Identify patterns

Common patterns:

• Spikes in 429: Need rate limit increase
• 401 errors: Key was revoked or blocked
• 402 errors: Wallet needs refill
• 500 errors: Contact support if persistent

Team Analytics (Admin/Owner)

Per-Member Usage

View individual member activity:

1. Go to Request Logs
2. Filter by team member
3. See their metrics:
   • Total requests
   • Token usage
   • Cost incurred
   • Models used
   • Error rates

Use cases:

• Identify power users
• Budget allocation
• Training needs
• Cost attribution

Per-Key Analytics

Track API key usage:

1. Go to Request Logs
2. Filter by API key
3. Analyze:
   • Usage patterns
   • Cost trends
   • Error rates
   • Model preferences

Use cases:

• Identify problematic keys
• Optimize key limits
• Debug integration issues
• Plan key rotation

Team Spending Trends

Monitor team costs:

• Daily spending breakdown
• Week-over-week trends
• Month-over-month growth
• Per-member cost attribution

Set team budgets:

• Organization-wide limits
• Per-key limits
• Per-member quotas (via key limits)
• Alert on threshold breaches

Export and Reporting

Data Export

Available formats:

• CSV (coming soon)
• JSON (coming soon)
• API access (coming soon)

What you can export:

• Request logs
• Cost reports
• Usage summaries
• Token breakdowns

Use cases:

• Financial reconciliation
• Custom analytics
• Business intelligence
• Compliance reporting

Scheduled Reports

Coming soon:

• Daily usage summaries
• Weekly cost reports
• Monthly spending analysis
• Custom report schedules

Access Control

Who Can View Analytics?

Privacy

Member privacy:

• Members see only own activity
• Cannot view team spending
• No access to other members' logs
• Perfect for contractors/consultants

Team transparency:

• Admins/Owners see all activity
• Required for cost management
• Enables proper budgeting
• Supports accountability

Best Practices

Regular Monitoring

Daily:

• Check wallet balance
• Review any errors
• Monitor spending trends

Weekly:

• Analyze per-key usage
• Review team activity
• Check for anomalies

Monthly:

• Full cost analysis
• Model usage review
• Budget vs actual comparison
• Optimize based on trends

Setting Up Alerts

Wallet alerts:

• Low balance threshold ($10 default)
• Email when triggered
• Adjust in Organization Settings

API key alerts:

• Monthly limit warnings (75%, 90%, 100%)
• Automatic blocking at limit
• Email notifications

Usage alerts (coming soon):

• Daily spending thresholds
• Unusual activity detection
• Error rate spikes

Cost Governance

For small teams:

• Set reasonable monthly limits per key
• Monitor weekly
• Review monthly for trends

For larger teams:

• Per-key limits strictly enforced
• Per-member quotas
• Approval process for high-limit keys
• Monthly cost reviews with team leads

Troubleshooting

Missing Logs

Problem: Don't see recent requests in logs

Possible causes:

• Filters applied (check filter settings)
• "My Logs Only" toggled (if you're looking for team logs)
• Page not refreshed (refresh to see latest)

Solution: Clear filters and refresh

Costs Don't Match Expectations

Problem: Costs higher than anticipated

Investigation steps:

1. Check Request Logs for high-cost requests
2. Identify which models being used (GPT-4 vs GPT-3.5)
3. Look at token counts (unexpectedly long outputs?)
4. Check for errors (retries can double costs)
5. Review API key usage (is a key being overused?)

Solutions:

• Switch to cheaper models where appropriate
• Reduce max_tokens parameter
• Optimize prompts to be more concise
• Fix error causes to prevent retries
• Set stricter API key limits

High Error Rates

Problem: Many failed requests

Check:

1. Filter logs by error status (4xx, 5xx)
2. Read error messages
3. Identify patterns (same key? same model? same time?)

Common fixes:

• 401: Check API key is valid
• 402: Refill wallet
• 429: Reduce request rate or contact for limit increase
• 500/503: Wait and retry, or contact support if persistent

Next Steps

• View detailed request logs
• Set up API key limits
• Monitor wallet balance
• Review audit logs