Trace model requests
Send spans around LLM calls and inspect their duration, status and attributes. Include provider and model names in your instrumentation to keep that context with the request.
AI monitoring
Monitor AI apps.
Monitor AI apps.
Find the slow step.
Trace model requests, retrieval and tool calls with OpenTelemetry. See where your AI application spends its time, inspect failures, and connect the request to its logs in Site Qwality.
Bring your OpenTelemetry instrumentation
Free telemetry tier
No credit card
Your AI request is part of a bigger system.
A slow answer might start with a model call, a database query or a tool request. Instrument those steps as spans and inspect them together, alongside the services your application already depends on.
Follow agent and RAG workflows
Keep retrieval, model calls and tool execution in the same trace by propagating trace context. The waterfall shows the steps your instrumentation captures.
Find slow or failed calls
Search traces by service, duration and status. Filter on the span attributes you send to narrow an investigation to a provider, model or operation.
Read the matching logs
Send application logs with the same trace ID and open them from the trace. Put a failed model request next to the application events that explain it.
Check your AI application API
Use HTTP monitors to check your application endpoints on a schedule. Configure requests and expected responses, then route availability alerts to your team.
Investigate through your assistant
Connect Claude, ChatGPT, Cursor or another remote MCP client to Site Qwality. Grant access to traces, logs and incidents so your assistant can help investigate. Explore the MCP server.
OpenTelemetry
Use the instrumentation
Use the instrumentation
your application controls.
Hosted models, a self-hosted model or a mix of both: send the spans your application emits through OTLP over HTTP. Your instrumentation determines which calls and attributes appear. Site Qwality does not need access to your model provider account.
- Instrument model calls, retrieval and tools with your OpenTelemetry SDK or compatible instrumentation
- Export traces over HTTP in protobuf or JSON
- Follow the OTLP setup guide for authentication and exporter settings
An instrumented app: trace exporter settingsOTEL_SERVICE_NAME=ai-api
OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=https://traces.siteqwality.com/v1/traces
OTEL_EXPORTER_OTLP_TRACES_PROTOCOL=http/protobuf
OTEL_EXPORTER_OTLP_HEADERS="Authorization=Bearer YOUR_API_KEY"
OTEL_METRICS_EXPORTER=none
OTEL_LOGS_EXPORTER=none
OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=https://traces.siteqwality.com/v1/traces
OTEL_EXPORTER_OTLP_TRACES_PROTOCOL=http/protobuf
OTEL_EXPORTER_OTLP_HEADERS="Authorization=Bearer YOUR_API_KEY"
OTEL_METRICS_EXPORTER=none
OTEL_LOGS_EXPORTER=none
One investigation
Start with the slow answer.
Start with the slow answer.
Follow it through your stack.
Open the request trace, find the span that took the time, and inspect its attributes and status. Read matching application logs to see what happened around the call. The same workflow works for an AI feature and the rest of your application.
- Compare the duration of retrieval, model requests and tool calls that you instrument
- Keep provider, model and request metadata on the spans you send
- Use MCP to investigate with your own AI assistant, with the permissions you choose
FAQ
You can trace calls to these providers when your application or compatible instrumentation emits OpenTelemetry spans for them. Site Qwality receives those spans over OTLP/HTTP. It does not automatically instrument a provider SDK when you sign up. Use the trace exporter guide to connect an instrumented application.
Yes, through the spans you send. Instrument retrieval, model calls and tool execution, and propagate trace context between steps to see them as one request. Steps without instrumentation will not appear in the trace.
This page covers request tracing, application logs, API availability and MCP access. It does not include automatic token-cost calculations, prompt management, answer evaluations or hallucination detection.
No. You choose which attributes your instrumentation exports. Duration, status, provider and model metadata can help investigate performance without sending prompt or response text. Review your instrumentation settings before sending application data.
AI monitoring helps you inspect the performance and failures of your own AI application. The MCP server lets an AI assistant read or manage Site Qwality data within the permissions you grant. You can use either one independently.
Yes. Logs, metrics and traces share the free telemetry allowance. MCP access is also included on the free plan. See pricing for current allowances and retention. Your model provider and AI assistant subscriptions are separate.
Related features
Start free. No credit card.
Uptime, cron, synthetic, logs, RUM, incidents, and status pages. Free tier on every product.