Grafana, Prometheus, Tempo and Loki
Guidely can show your own backend metrics, logs and traces next to its errors, traces, session replays, releases and incidents. It queries your Grafana, Prometheus, Tempo or Loki live when a page opens, and never stores what comes back.
So when an error happens, you can see in the same place:
- the backend trace it belongs to, span by span, from Tempo;
- the log lines that mention its trace id, from Loki;
- request rate, 5xx rate and latency for the service around that moment, from Prometheus.
Where the panels appear
| Page | What it shows |
|---|---|
| Error event (Monitor → Errors → an event) | The backend trace for the event's trace id and the log lines that mention it. |
| Trace (Monitor → Performance → a trace) | The backend spans next to the browser's spans, and the matching log lines. |
| Session replay | Log lines for the trace ids of the requests made during the recording, and the backend trace of a request you pick. |
| Release | Metric panels around the deploy, with the deploy marked on the chart. |
| Uptime incident | Metric panels and error log lines from the minutes before and during the incident. |
| Metric alert incident | The same, around the time the alert fired. |
| AI issue analysis | When it runs, it can read the backend trace and the log lines for the error, so its explanation covers what the server did. |
| AI assistants (MCP) | get_backend_trace, query_backend_logs, query_backend_metrics and search_backend_traces (MCP tools). |
Panels only appear in projects that have a connection set up (see project setup) and only for people with the View backend traces, logs and metrics permission (observability.view). Every role has it by default; admins can revoke it for individual members in Settings → Team.
Nothing is stored
Each panel is a live, read-only query made when the page opens. The answer goes back to your browser and is not written to Guidely's database, logs, caches or backups. Guidely stores only the connection settings: URLs, data source ids, the per-environment selectors and panel queries, and (for direct connections) the credentials, encrypted. Opening the same page twice queries your backend twice.
Live queries are rate limited to 300 a minute per workspace, and a panel can cover at most 7 days. More on what is sent where is in Security.
Choosing how to connect
There are two independent choices for each connection.
What Guidely queries
| Grafana | Prometheus, Tempo and Loki directly | |
|---|---|---|
| Setup | One service account token with the Viewer role; pick a Prometheus, Loki and Tempo data source | One URL (and credentials) per backend |
| Credentials | One token, for Grafana | One per backend, as each needs |
| "Open in Grafana" links | Yes, to Explore with the same query | When you give your Grafana's URL and the data source UID |
| Good for | Grafana Cloud, or a self-hosted Grafana in front of everything | Grafana Cloud stack URLs, Mimir, a single Prometheus |
See Connect Grafana and Connect Prometheus, Tempo or Loki directly.
How Guidely reaches it
| Direct | Relay | |
|---|---|---|
| Network | Guidely calls a public URL (Grafana Cloud, or your own behind TLS) | A small agent in your network connects out to Guidely; nothing connects in |
| Credentials | Stored in Guidely, encrypted | Stay in the relay's environment; Guidely never sees them |
| Private addresses | Refused | Yes: anything the relay can reach |
| Runs on | Nothing to run | Docker, Kubernetes, AWS ECS Fargate or Google Cloud Run |
Use direct for anything already reachable from the internet. Use the relay for a stack on a private network, or when credentials must not leave your environment. See The relay.
You can mix them: for example Grafana Cloud directly for metrics, and Loki through a relay.
Setting it up
- A workspace admin adds a connection in Integrations → Grafana, Prometheus, Tempo, Loki (and, for private stacks, a relay first). Test checks it before saving.
- Someone with the Developer settings permission picks, per project, which connection answers metrics, logs and traces, and maps each environment onto your labels. See project setup.
Trace ids that match
The panels line up because Guidely and your backend use the same trace ids.
- The Guidely web SDK, with performance tracing on, sends a W3C
traceparentheader (andsentry-trace/baggage) on the requests it traces. A backend instrumented with OpenTelemetry (or a Sentry or Guidely server SDK) continues that trace, so the browser's trace id is the one in Tempo and in your logs. Performance monitoring has the details. - By default only same-origin requests get the headers. For an API on another origin, list it in
tracePropagationTargetsand allow thetraceparent(andsentry-trace,baggage) request headers in that API's CORSAccess-Control-Allow-Headers; otherwise the browser's preflight fails.createGuidely({ projectId: "pk_…", tracing: { tracePropagationTargets: ["api.example.com", /^\/api\//] }, });
- Errors captured by Sentry SDKs on your server carry the trace id of the request they happened in, so they match too.
- For log lines to match, your logs need the trace id: either in the line itself (most OpenTelemetry log bridges and logging libraries' trace correlation write
trace_id=…or"traceId":"…"), or as a Loki label or structured metadata fieldtrace_id. Project setup explains the two options.