Skip to main content

Monitoring and Telemetry Integration

Radium automatically tracks token usage, costs, and execution metrics during plan execution. This data helps you understand resource consumption and optimize your workflows.

Overview​

Monitoring integration provides:

  • Token Usage Tracking: Prompt and completion tokens per task
  • Cost Calculation: Estimated costs based on engine and model pricing
  • Session Analytics: Aggregated metrics across all tasks
  • Agent Tracking: Per-agent execution metrics

Automatic Telemetry​

Telemetry is recorded automatically for each task execution:

# Execute plan with monitoring
rad craft

# Output includes token usage:
# → Executing I1.T1...
#   • Tokens: 1500 prompt, 800 completion
#   • Progress: 20%

Token Usage​

Token usage is tracked from model responses:

  • Input Tokens: Tokens in the prompt sent to the model
  • Output Tokens: Tokens in the model's response
  • Total Tokens: Sum of input and output tokens

Example​

Task: I1.T1
  Input Tokens: 1,500
  Output Tokens: 800
  Total Tokens: 2,300

Cost Calculation​

Costs are calculated automatically based on:

  • Engine: OpenAI, Claude, Gemini, etc.
  • Model: Specific model used (gpt-4, claude-3-sonnet, etc.)
  • Token Usage: Input and output tokens

Pricing Examples​

OpenAI GPT-4:

  • Input: $30 per 1M tokens
  • Output: $60 per 1M tokens

Claude 3 Sonnet:

  • Input: $3 per 1M tokens
  • Output: $15 per 1M tokens

Claude 3 Haiku:

  • Input: $0.25 per 1M tokens
  • Output: $1.25 per 1M tokens

Cost Calculation Example​

# Task uses Claude 3 Sonnet
# Input: 1,000 tokens
# Output: 500 tokens

# Cost calculation:
# Input: (1,000 / 1,000,000) * $3 = $0.003
# Output: (500 / 1,000,000) * $15 = $0.0075
# Total: $0.0105

Session Analytics​

After plan execution, session analytics are generated:

# Execute plan
rad craft

# Session report (if monitoring available):
# Session: REQ-123
# Total Tokens: 45,000
# Total Cost: $0.45
# Execution Time: 0:15:30
# Agents Used: code-agent, review-agent

Engine and Model Tracking​

Telemetry records which engine and model were used:

# Telemetry includes:
# - Engine ID: claude
# - Model: claude-3-sonnet
# - Provider: claude

This helps track costs and performance across different providers.

Monitoring Service​

Monitoring data is stored in .radium/monitoring.db:

# Monitoring database location
.radium/monitoring.db

The monitoring service:

  • Stores telemetry records per task
  • Tracks agent execution status
  • Generates session reports
  • Provides historical analytics

Graceful Degradation​

If monitoring is unavailable, execution continues without telemetry:

# If monitoring.db cannot be created:
rad craft

# Output:
# ⚠ Warning: Failed to record telemetry: <error>
# → Execution continues normally

Viewing Telemetry​

Use the monitoring commands to view telemetry:

# View session metrics
rad monitor session REQ-123

# View agent metrics
rad monitor agent code-agent

# View cost summary
rad stats

Best Practices​

  1. Monitor Costs: Regularly check token usage and costs
  2. Optimize Models: Use cheaper models when appropriate
  3. Track Trends: Monitor token usage trends over time
  4. Set Budgets: Use budget tracking to control spending
  5. Review Analytics: Review session analytics to optimize workflows

Cost Optimization​

Use Efficient Models​

# For simple tasks, use cheaper models
rad craft --engine claude --model claude-3-haiku

# For complex tasks, use more capable models
rad craft --engine claude --model claude-3-opus

Monitor Token Usage​

# Check token usage after execution
rad stats

# Output:
# Total Tokens: 45,000
# Total Cost: $0.45
# Average per Task: 2,250 tokens

Set Budget Limits​

# Set daily budget limit
rad budget set --daily 10.00

# Check budget status
rad budget status

Troubleshooting​

Telemetry Not Recording​

  • Check .radium/monitoring.db exists and is writable
  • Verify monitoring service initialized successfully
  • Check for permission errors in workspace directory

Incorrect Costs​

  • Verify engine and model are correctly identified
  • Check pricing is up-to-date for your model
  • Review token counts for accuracy

Missing Session Data​

  • Ensure monitoring service is available
  • Check that execution completed successfully
  • Verify session ID is correctly tracked

Telemetry Record Structure​

Each telemetry record includes:

{
  "agent_id": "code-agent",
  "timestamp": 1234567890,
  "input_tokens": 1500,
  "output_tokens": 800,
  "total_tokens": 2300,
  "estimated_cost": 0.0105,
  "model": "claude-3-sonnet",
  "provider": "claude",
  "engine_id": "claude"
}

See Also​