> ## Documentation Index
> Fetch the complete documentation index at: https://prismeai-docs-next.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Observability

> Monitor workspace performance, errors, costs, and request traces

<Frame>
  <img src="https://mintcdn.com/prismeai-docs-next/v9QCFXfb2pw6kaoq/images/ai-governance-observability.png?fit=max&auto=format&n=v9QCFXfb2pw6kaoq&q=85&s=eb5c66520cf1194dc52579b09e6a0c0a" alt="Observability Dashboard" width="1440" height="900" data-path="images/ai-governance-observability.png" />
</Frame>

Observability provides deep visibility into your AI workspaces, tracking performance metrics, error rates, LLM costs, and detailed request traces. Access it from **Observability** in the Governe sidebar.

## Dashboard Overview

The observability dashboard shows key metrics at a glance:

| Metric               | Description                                                      |
| -------------------- | ---------------------------------------------------------------- |
| **Health Score**     | 0-100 composite score based on errors, latency, and availability |
| **Total Executions** | Number of automation runs in the selected period                 |
| **Error Rate**       | Percentage of executions that failed                             |
| **P95 Latency**      | 95th percentile response time                                    |
| **LLM Costs**        | Total cost of LLM calls                                          |

## Health Score

The health score combines multiple factors into a single 0-100 metric:

| Status       | Score Range | Color  |
| ------------ | ----------- | ------ |
| **Healthy**  | 90-100      | Green  |
| **Degraded** | 70-89       | Yellow |
| **Critical** | 0-69        | Red    |

### Factor Breakdown

| Factor           | Weight | Calculation                     |
| ---------------- | ------ | ------------------------------- |
| **Error Rate**   | 40%    | Lower error rate = higher score |
| **Latency**      | 30%    | Lower latency = higher score    |
| **Availability** | 30%    | Higher uptime = higher score    |

## Performance Metrics

### Latency Percentiles

Track response time distribution:

| Percentile | Description                      |
| ---------- | -------------------------------- |
| **P50**    | Median response time             |
| **P90**    | 90% of requests faster than this |
| **P95**    | 95% of requests faster than this |
| **P99**    | 99% of requests faster than this |

### Slowest Automations

Identify bottlenecks by viewing automations ranked by average execution time.

### Timeline View

Visualize performance over time with:

* P95 latency trend (area chart)
* Request volume (line overlay)

This helps correlate load spikes with latency increases.

## Error Tracking

### Error Distribution

View errors grouped by:

* **Type**: `InvalidExpressionSyntax`, `FetchError`, `TimeoutError`, etc.
* **Automation**: Which automations generate the most errors

### Error Rate Trend

Track error rate changes over time to identify:

* New bugs introduced
* External service outages
* Configuration issues

### Recent Errors

View the latest errors with:

* Timestamp
* Error type and message
* Automation name
* Correlation ID (for tracing)

## LLM Costs

Monitor AI model consumption and costs:

### Cost Breakdown

| Metric            | Description                    |
| ----------------- | ------------------------------ |
| **Total Calls**   | Number of LLM API calls        |
| **Total Cost**    | Estimated cost in USD          |
| **Total Tokens**  | Input + output tokens consumed |
| **Carbon Impact** | Estimated CO2 emissions (kg)   |

### By Model

See which models consume the most resources:

* Model name
* Number of calls
* Cost percentage
* Token count

### By Automation

Identify which automations drive LLM costs:

* Automation name
* Call count
* Relative usage

### Cost Timeline

Track daily cost trends to:

* Forecast monthly spend
* Identify cost anomalies
* Plan capacity

## Request Traces

Trace individual requests through the entire execution flow.

### Accessing Traces

1. From the Observability dashboard, click **View Traces**
2. Or click a correlation ID in the error list

### Trace View

Each trace shows:

| Element         | Description                      |
| --------------- | -------------------------------- |
| **Summary**     | Event count, duration, user info |
| **Timeline**    | Waterfall view of all events     |
| **Automations** | List of automations executed     |
| **Errors**      | Any errors encountered           |

### Timeline Visualization

The waterfall chart shows:

* Each automation as a horizontal bar
* Start time offset from request start
* Duration of each step
* Errors highlighted in red

## Dependency Graph

Visualize how automations call each other.

### Graph View

* **Nodes**: Automations and installed apps
* **Edges**: Calls between components
* **Edge weight**: Call frequency

### Use Cases

* Understand system architecture
* Identify critical paths
* Find orphaned automations

## Anomaly Detection

Automatic detection of unusual patterns.

### Anomaly Types

| Type              | Description                        |
| ----------------- | ---------------------------------- |
| **Error Spike**   | Sudden increase in error rate      |
| **Latency Spike** | Sudden increase in response time   |
| **Traffic Spike** | Unusual increase in request volume |
| **Traffic Drop**  | Unusual decrease in activity       |

### Severity Levels

| Severity     | Threshold   |
| ------------ | ----------- |
| **Warning**  | 2x baseline |
| **Critical** | 3x baseline |

### Configuration

Adjust detection sensitivity:

* **Low**: Only alert on significant anomalies
* **Medium**: Balanced detection (default)
* **High**: Alert on minor deviations

## Period Comparison

Compare metrics between two time periods.

### How to Compare

1. Select current period dates
2. Select comparison period dates
3. View side-by-side metrics

### Compared Metrics

* Execution count
* Error count and rate
* Average and P95 latency
* Unique users

### Trend Indicators

* **Up**: Increase > 5%
* **Down**: Decrease > 5%
* **Stable**: Within ±5%

## Real-Time Feed

Monitor events as they happen.

### Event Types

| Type                           | Description          |
| ------------------------------ | -------------------- |
| `runtime.automations.executed` | Automation completed |
| `runtime.fetch.completed`      | HTTP fetch succeeded |
| `runtime.fetch.failed`         | HTTP fetch failed    |
| `runtime.llm.completed`        | LLM call completed   |
| `error`                        | General error event  |

### Filtering

* **All**: Show all events
* **Errors**: Only error events
* **Executions**: Only completion events

### Time Window

Configure how far back to show events (default: 60 minutes).

## Time Range Selection

All views support time range selection:

| Period     | Description         |
| ---------- | ------------------- |
| **1h**     | Last hour           |
| **24h**    | Last 24 hours       |
| **7d**     | Last 7 days         |
| **30d**    | Last 30 days        |
| **Custom** | Specific date range |

## Platform Summary (Admin Only)

Platform administrators see a platform-wide summary:

* Aggregate health score across all workspaces
* Total platform requests
* Combined LLM costs
* Peak capacity usage

## Best Practices

<CardGroup cols={2}>
  <Card title="Set Baselines" icon="chart-line">
    Establish normal performance baselines for comparison
  </Card>

  <Card title="Monitor Anomalies" icon="bell">
    Enable anomaly detection for early warning
  </Card>

  <Card title="Track Costs" icon="dollar-sign">
    Review LLM costs weekly to prevent surprises
  </Card>

  <Card title="Use Traces" icon="magnifying-glass">
    Investigate errors using correlation IDs
  </Card>
</CardGroup>

## Troubleshooting

<Tabs>
  <Tab title="High Latency">
    1. Check **Slowest Automations** for bottlenecks
    2. Review the **Timeline** for load correlation
    3. Examine traces for slow steps
    4. Check external API response times
  </Tab>

  <Tab title="High Error Rate">
    1. View **Error Distribution** by type
    2. Check **Recent Errors** for patterns
    3. Trace errors using correlation IDs
    4. Review dependency health
  </Tab>

  <Tab title="Cost Spikes">
    1. Check **By Model** breakdown
    2. Identify expensive automations
    3. Review **Cost Timeline** for anomalies
    4. Check for loops or inefficient prompts
  </Tab>
</Tabs>

## Next Steps

<CardGroup cols={2}>
  <Card title="Model Governance" icon="robot" href="./model-governance">
    Control model access and costs
  </Card>

  <Card title="Audit Logs" icon="scroll" href="./audit-subscriptions">
    Track administrative changes
  </Card>
</CardGroup>