Cost Intelligence

Cost Intelligence analyzes your LLM spend and provides actionable recommendations to reduce costs without sacrificing quality.

Insights

The analyzer generates suggestions based on your actual usage patterns:

Suggestion Type	Description
model_swap	Switch to a cheaper model for simple tasks
caching	Enable prompt caching for repeated patterns
token_reduction	Reduce prompt length or context window
batching	Batch similar requests together
timeout	Add timeouts to prevent runaway sessions
budget_guard	Set hard budget caps per session

Per-Session Analysis

Break down costs at the step level:

curl -H "Authorization: Bearer $TOKEN" \
  https://api.caged.dev/v1/cost-intel/sessions/{session-id}

Returns:

• Step-by-step cost breakdown
• Model used per step
• Token counts (input/output)
• Hotspot detection (which steps cost the most)
• Optimization suggestions

Model Comparison

Compare cost/quality across providers:

curl -H "Authorization: Bearer $TOKEN" \
  "https://api.caged.dev/v1/cost-intel/models?period=7d"

Returns metrics per model:

• Total cost
• Total tokens
• Average cost per 1K tokens
• Call count
• Suggested swaps

Daily Trends

curl -H "Authorization: Bearer $TOKEN" \
  "https://api.caged.dev/v1/cost-intel/insights?period=week"

Supported Models

Built-in pricing for:

• Claude Sonnet 4 ($3/$15 per MTok)
• Claude Haiku ($0.25/$1.25 per MTok)
• GPT-4o ($2.50/$10 per MTok)
• GPT-4o-mini ($0.15/$0.60 per MTok)
• Claude Opus 4 ($15/$75 per MTok)

Dashboard

View cost analytics at /dashboard/costs:

• Daily spend charts
• Per-model breakdown
• Top spending sessions
• Optimization suggestions