Logs and Metrics

The MultiRoute dashboard provides request-level logs and aggregate metrics to help you understand how your applications are using the /v1 API and how underlying providers are behaving.

Even while observability features are in beta, the intent is to give you a clear, debuggable picture of:

• What requests are being made
• How long they take
• Where errors are coming from

Request Logs

The Logs view surfaces recent requests grouped by project and key. Each log entry typically includes:

• Timestamp: When the request was received.
• Request ID: A unique ID you can use to correlate with your own logs.
• Endpoint: For example, /v1/chat/completions.
• Model and provider: Which underlying model MultiRoute routed the request to.
• Status: Success, provider error, or client error.
• Latency: End-to-end response time as seen by MultiRoute.

You can filter or search logs by:

• Time range (e.g. last 15 minutes, 24 hours).
• API key or project.
• Endpoint, status, or model.

These filters are especially useful when debugging specific incidents or deployments.

Latency Metrics

The Latency section focuses on performance over time. Typical views include:

• p50, p95, p99 latencies per endpoint, provider, and model.
• Trend charts over time windows (e.g. last hour, last day).
• Comparisons between models or routing strategies within the same project.

Use these metrics to:

• Spot regressions after configuration or model changes.
• Choose models that offer the right balance of cost and latency.
• Identify providers that are intermittently slow.

Error Metrics

The Errors view aggregates failures so you can quickly understand:

• Overall error rate for a project or endpoint.
• Breakdown by provider and model.
• Common error types, such as provider timeouts, quota issues, or invalid requests.

This helps distinguish:

• Client-side issues (invalid parameters, authentication failures) that you can fix in your code.
• Provider-side issues (outages, rate limits, timeouts) that MultiRoute may automatically mitigate via retries or failover.

Correlating Logs with Your Application

Each request is assigned a request ID and may also include a correlation ID if you pass one in headers. In your application:

• Log the MultiRoute request ID whenever you make a call.
• Optionally provide your own correlation ID (for example, per user or per job) and use it in both your logs and MultiRoute requests.

With this in place you can:

• Start from an error or slow response in your system.
• Look up the matching request in the dashboard.
• Inspect which provider and model were used and what errors or latencies were observed.

Exporting and Integrating Metrics (In Progress)

The dashboard is designed to integrate with external observability tools via:

• Structured logs that can be shipped to your logging stack.
• Metrics export (such as Prometheus-style metrics) for use in your monitoring system.

These integrations are in progress. The recommended pattern is to:

• Use the dashboard for day-to-day debugging and inspection.
• Forward logs and metrics from your infrastructure to your existing APM/monitoring tools for long-term storage and alerting.