Observability in AI Management shows the traffic that passes through AI Gateway: calls to models, MCP servers, and agents. In the Konnect sidebar, click Observability to open these pages:
|
Page |
Use it to |
|---|---|
| Overview | Review the health and cost of AI traffic. |
| Dashboards | Build and share your own dashboards from AI Gateway metrics. |
| Explorer | Answer specific questions with custom queries. |
| Logs | Read and search logs. |
To narrow the Overview page, use the following filters:
To change the period, use the time range menu. The default period is the last 7 days.
The page has the following sections. Click a section name at the top of the page to jump to it.
|
Section |
Charts |
|---|---|
| Key metrics | LLM latency (p95), Request count, Total token count, Cached token count, Error rate, and Costs, each compared with the previous period |
| Top consumers | Models, Providers, Users, and MCP Server |
| Performance | Total requests, LLM latency, Error rate, Tokens per minute, Time to first token (TTFT), and Time per output token (TPOT) |
| Costs | Total costs vs request count and Cost by model |
| Models | Success rate and usage by model: Model usage by user, Error rate per model, and Latency per model |
| MCP | Tool usage by model across MCP servers: Method calls by model and Failing method calls |
The LLM latency, Time to first token (TTFT), and Time per output token (TPOT) charts show p50, p95, and p99.
The data you see in Observability depends on your AI Management role:
Use dashboards to view AI Gateway metrics. Click Manage dashboards to see the dashboards in your organization. To add one, click Create dashboard, then create it from a template, from scratch, or from a JSON definition.
The following templates are built for AI traffic:
|
Template |
Description |
|---|---|
| AI gateway dashboard | AI-focused insight into your LLM traffic and operational costs. Track request volume by model and provider, understand token usage trends, and monitor latency across different models. Teams operating AI Gateways can use this dashboard to optimize performance and manage costs by monitoring traffic patterns and provider-specific behavior. |
| Agentic analytics dashboard | Overview of agentic traffic flowing through your AI Gateway. This dashboard highlights which tools are called most frequently, breaks down tool usage by consumer, and tracks average latency per tool over time. It helps teams operating MCP-enabled services understand usage patterns and identify performance bottlenecks. |
To see cost next to latency or TTFT, add a tile for each metric to the same dashboard.
The Explorer in AI Management works the same as the Konnect Explorer. For the full Explorer documentation, including time ranges, saving reports, and exporting to CSV, see the Konnect Explorer reference.
In the Explorer query bar, you can build a query From a data source, Show a chart type with a metric, per a time interval, by a dimension.
For AI traffic, use one of these data sources:
|
Data source |
Metrics |
|---|---|
| LLM usage |
|
| Agentic usage |
|
To group or filter AI traffic, use these dimensions:
| Dimension | LLM usage | Agentic usage |
|---|---|---|
| AI gateway | Supported | Supported |
| AI gateway model | Supported | Supported |
| AI gateway MCP server | Supported | Supported |
| AI gateway agent | Supported | Supported |
| AI gateway consumer | Supported | Supported |
| AI gateway consumer group | Supported | Supported |
| Provider | Supported | Supported |
| Principal | Supported | Supported |
| Principal group | Supported | Supported |
| Status code (grouped) | Supported | Supported |
| Request model | Supported | Not supported |
| Response model | Supported | Not supported |
| MCP tool name | Not supported | Supported |
| MCP method | Not supported | Supported |
| MCP error | Not supported | Supported |
The Logs page shows your AI Gateway data plane and control plane logs, newest first.
How long is Observability data kept?
The metrics that Overview, Dashboards, and Explorer show are kept for one year.