AI Observability Overview
The overview dashboard provides a consolidated view of AI services usage in Amorphic, covering token consumption, request volumes, response times, and guardrail activity. KPI cards and charts help assess system behavior at a glance.
KPI cards and charts are displayed to understand system behavior at a glance.
AI Observability is available only when AI services are enabled in the Amorphic platform.
Key Terms
- Token: The basic unit of text (words or characters) processed by AI models.
- Model: The AI engine processing your requests (e.g., Claude 3.5 Sonnet).
- Catalog: The semantic search system used to find and retrieve data.
- Latency: The time it takes to process a request, measured in milliseconds.
- Guardrail: Safety filters that catch and block sensitive or inappropriate content.
- Service: Amorphic features that use AI models (e.g., Agents, Data Pipelines).
Getting Started
- Navigate to the AI Observability tab
- Go to the Overview tab
- Check the Data Refresh timestamp
- Explore KPI Cards and Charts
Data Refresh
The page shows a Data as of timestamp at the top right of the page that reflects when the dashboard data was last synchronized. Data syncs automatically every hour. Click Refresh to load the most recent available data.
KPI Cards
The KPI cards at the top of the page provide a quick summary of AI activity for the last 7 days. Each card includes:
- A rolled-up value for the 7-day window
- A small daily trend sparkline
- Percentage change compared to the previous 7-day period
Total Tokens: Total token consumption across all AI services. Use this to track overall AI usage and spot consumption trends over time.
Average Catalog Search Latency: Average response time for catalog searches. Monitor search performance and detect latency spikes.
Total Requests: Total number of requests made to AI-powered services. Useful for understanding request volume trends and overall system activity.
Average Guardrail Hit Rate: Percentage of requests blocked by guardrails. Helps monitor content filtering effectiveness and safety enforcement trends over time.
Charts
The charts provide a detailed breakdown of recent AI activity, highlighting the most active entities based on the latest usage patterns, along with their trends over time.
Tokens Used by Service — Daily token usage over the last 7 days for the top 5 services by consumption. Helps identify which services drive the highest AI usage.
Token Distribution by Model — Token consumption share across the top 5 models in the last hour. Helps understand how usage is distributed across different models.
Top Token Consumers — Top 5 users by token consumption in the last hour, listed in descending order. Helps monitor user-level resource usage.
Average Latency by Model — Average latency trends over the last 7 days for the top 5 models ranked by latency in the last hour. Helps identify performance patterns and latency spikes.
Top Requests by User — Daily request counts for the top 5 users ranked by service consumption in the last hour. Helps understand system usage and engagement patterns.
Guardrail Hit Rate — Daily percentage of requests blocked for the top 5 guardrails ranked by hit rate in the last hour. Helps monitor content filtering effectiveness and safety activity over time.
Next Step
- Use the KPI Explorer tab for detailed time-series analysis with custom filtering