Organization
The Dashboard shows the Gateway traffic of every project and application you can access. It tells you how much your organization sends, what it costs, and where requests fail or get blocked.
Required role
Applies to
Why use it
Use the Dashboard when you need one view across teams, such as a cost review or a check after an incident. Each section helps with questions such as:
| Section | Use it to find out |
|---|---|
| Usage | How much traffic goes through the Gateway, and how many tokens does it use? |
| Cost | How much do you spend, and what made it go up or down? |
| Reliability | How many requests succeed, how fast are they, and why do the others fail? |
| Governance & Security | How often do guardrails run, block or detect something, and in which applications? |
| Data Integrations | How much does retrieval add to requests, and how often does it fail? |
| Tools | How many tool calls do applications make, and how many succeed? |
How it works
In the console, open Dashboard in the sidebar. The Dashboard contains information about every LLM request and MCP tool call that applications you can see send through the Gateway. It opens on the last 7 days. At the top right, pick Today, 7d, 30d or Custom. The period applies to the whole page.
Every section has the same three parts:
Cost
Spend
$1,284.50 +12.4%
Average spend per day
$183.50 +12.4%
Average cost per request
$0.0214 -3.1%
Blended price per million tokens
$3.12 -2.0%
- Summary cards. Each card shows one metric for the period and its change against the previous period. The previous period is the same length of time, immediately before. Select the info icon on a card to read the definition of the metric.
- A chart. Above the chart, pick a Metric, a Grouping and a Bucket. The grouping splits the vertical bars by categories the metric allows, like application, project, provider, model, reason or tool. The bucket is the time step of the chart's x-axis, such as one day.
- A table. Switch the chart to a table to compare, for each category of the grouping, the aggregated metric over the period with the previous period. For most metrics, the table also shows how much of the change each category caused. For a metric that is a sum, the rows of Share of change add up to the Change % of the total.
In a chart with daily buckets, select a day to show only that day. Clear it to go back to the full period. In a table grouped by application, select an application row to open that application.
Dashboard sections
Each section links to the definitions of its metrics in Metrics.
Usage
Usage shows how much traffic goes through the Gateway: requests and tokens.
- When input tokens rise faster than requests, the average request sends more input tokens.
- A higher cache hit rate is better. Group it by model to find the models that read less from the prompt cache.
Cost
Cost shows what your traffic costs, at the price of each model: spend and its rates.
- When spend rises, open the table. It splits the change into volume, mix and rate. See How a change splits into volume, mix and rate.
- For average cost per request and the blended price, lower is better. When the blended price falls but spend rises, you used more tokens.
Reliability
Reliability shows whether requests succeed and how long they take: outcomes and latency.
- Group the success rate by provider or model to find one that fails more than the others.
- When p95 latency is far above p50, a small number of requests are slow, not all of them.
- Use Top failure reasons to see what fails most.
Governance & Security
This section shows what your guardrails do with your traffic: guardrail activity.
- One request can run several guardrail checks, so evaluations can be more than requests.
- A high audit detection rate with a low blocked rate means guardrails find issues but block few of them. Review the action of each check in Guardrails.
- Use Violations by application to find the applications that trigger guardrails most often.
Data Integrations
Data Integrations shows what retrieval adds to requests: retrieval metrics.
- These metrics are empty when no request in the period used retrieval.
- Compare retrieval added latency with end-to-end latency in Reliability to see how much of the wait comes from retrieval.
- To connect a data source, see Data Integrations.
Tools
Tools shows the tool calls that go through the MCP Gateway: tool call metrics.
- Group by tool to find the tools that fail or run slowly.
- Group by application to see which applications depend on each tool.
Related
- Metrics: the definition and formula of every number on this page.
- Application: the same sections for one application, and its requests.
- Tools: the MCP servers behind the Tools section.