This dashboard provides a comprehensive view of your Open WebUI deployment using trace and log data. It combines two telemetry sources: Open WebUI's native OpenTelemetry auto-instrumentation of FastAPI, SQLAlchemy, and HTTP clients (server spans carrying http.route, database spans carrying db.system and db.operation), and OpenLIT GenAI instrumentation for model calls (chat spans carrying the OpenTelemetry gen_ai.* semantic conventions such as gen_ai.request.model, gen_ai.usage.input_tokens, and gen_ai.usage.output_tokens, plus OpenLIT's own gen_ai.usage.cost extension). Every panel keys off these attributes and filters through the service_name picker at the top, so you can scope the view to one or more services (for example the application service and the OpenLIT LLM service).
Dashboard Preview
Dashboards → + New dashboard → Import JSON
What This Dashboard Monitors
This dashboard tracks the performance, cost, and reliability of your Open WebUI deployment using OpenTelemetry trace and log data to help you:
- Track Application Health: See total request volume, error rate, and p95/p99 latency for the Open WebUI backend at a glance.
- Analyze Endpoints: Break down traffic and latency by
http.routeto find the busiest and slowest API endpoints. - Monitor the Database: Watch p95/p99 query latency and per-operation counts to catch database-driven slowness.
- Measure LLM Usage: Track model request volume, input and output tokens, and total cost across every model in use.
- Control LLM Cost: Follow spend over time and compare cost per model to keep the bill predictable.
- Watch LLM Latency: See p50, p95, and p99 latency for model calls to surface slow responses and regressions.
- Catch Errors Early: Watch the error rate by service and drill into recent failed traces and logs the moment incidents begin.
Panels Included
Application Health
| Panel | Type | What It Shows |
|---|---|---|
| Total Requests | Value | Count of inbound HTTP requests (FastAPI server spans) handled by the selected services |
| Error Rate | Value | Fraction of server requests that ended in an error status, shown as a percentage |
| Latency p95 | Value | 95th-percentile request duration across all endpoints |
| Latency p99 | Value | 99th-percentile request duration across all endpoints |
- Request Rate by Endpoint: Time-series graph of request volume grouped by
http.route, showing the traffic mix and which endpoints dominate. - Latency Percentiles (p50 / p95 / p99): Time-series graph of request duration percentiles, showing both typical performance and tail latency over time.
- Error Rate Over Time: Time-series graph of the error fraction of server requests, revealing incident windows and regressions.
- Request Distribution by Endpoint: Pie chart of each endpoint's share of requests (
http.route), a clear view of where traffic concentrates. - Endpoint Performance: Table of endpoints with request count and p95/p99 latency, for a side-by-side comparison to find the slow routes.
Database
- DB Query Latency (p95 / p99): Time-series graph of p95 and p99 duration for database (SQLAlchemy) query spans, surfacing database-driven slowness.
- Database Operations: Table of database operations grouped by
db.operation, with query count and average latency, showing which queries run most and which are slowest.
LLM / GenAI
| Panel | Type | What It Shows |
|---|---|---|
| LLM Requests | Value | Count of LLM chat requests captured by the OpenLIT gen_ai spans |
| Input Tokens | Value | Total prompt tokens consumed across LLM calls (gen_ai.usage.input_tokens) |
| Output Tokens | Value | Total completion tokens generated across LLM calls (gen_ai.usage.output_tokens) |
| Total Cost (USD) | Value | Summed OpenLIT-computed cost of LLM calls (gen_ai.usage.cost) |
- LLM Latency Percentiles (p50 / p95 / p99): Time-series graph of duration percentiles for LLM chat spans, showing model-call latency both typical and tail.
- Token Usage Over Time: Time-series graph of input vs output token volume, showing demand and how prompt/response sizes trend.
- LLM Cost Over Time (USD): Time-series graph of LLM spend, to spot cost spikes and trends before they surprise the bill.
- Requests by Model: Pie chart of each model's share of LLM requests (
gen_ai.request.model), showing which models carry the load. - Model Breakdown: Table of models with request count, input/output tokens, cost, and average latency, for the cost and usage comparison across models.
Errors & Logs
- Errors Over Time by Service: Time-series graph of error span count grouped by
service.name, pinpointing which service is failing and when. - Recent Error Traces: List of the latest errored spans across the selected services, for jumping straight to failures when the error rate climbs.
- Recent Logs: List of the most recent logs from the selected services, for quick log context to read alongside the traces.
