The managed LGTM stack, run for you in Germany
Loki for logs, Grafana to look at everything, Tempo for traces and Mimir for metrics. You get a Grafana of your own with the three stores behind it; we host, run and update all of it.
What you get
Four open-source parts, and a Grafana of your own.
Open-source Grafana, Mimir, Loki and Tempo, queried with PromQL, LogQL and TraceQL, and fed by standard agents: Grafana Alloy, OTLP over HTTP, Prometheus remote write and the Loki push API.
Loki
Logs
Takes your logs over the Loki push API or OTLP, and answers LogQL in Explore and on your dashboards. Alert rules on your logs run in Loki itself.
Grafana
Dashboards, Explore, alerting
One Grafana per customer, named after your organization. You sign in with your Telopi account, with no second password, and your role in Telopi sets your role in Grafana. Dashboards for uptime, machines and Kubernetes are built in.
Tempo
Traces
Takes your traces over OTLP. From your spans it writes service graphs and span metrics into your own Mimir, so calls, errors and latency per service are there to graph and alert on.
Mimir
Metrics
Stores your Prometheus metrics, sent by remote write or OTLP, and answers PromQL from your Grafana. Alert rules on your metrics run in Mimir itself.
How data gets in
Send with Grafana Alloy, OTLP or Prometheus remote write.
Your Grafana has one HTTPS address to send to, ingest.telopi.com followed by its name. Each signal has its path below it, and every sender except the website snippet signs in with a username and password.
Metrics
Prometheus remote write, or OTLP over HTTP
/mimir/api/v1/push/mimir/otlp/v1/metricsLogs
Loki push API, or OTLP over HTTP
/loki/loki/api/v1/push/loki/otlp/v1/logsTraces
OTLP over HTTP
/otlp/v1/traces
Grafana Alloy
The one thing to run on your side. Our Alloy configuration sends metrics, logs and traces with three settings, and your applications send OTLP to it over gRPC (4317) or HTTP (4318).
OTLP over HTTP
OTLP sent straight to us goes over HTTP; there is no gRPC ingest. An application that speaks only gRPC sends to Alloy, which passes it on.
Ingest keys
Up to five named keys per Grafana, each with a username of its own. A key can write and cannot read. Replace, rotate or disable one in the portal; it takes effect within about a minute.
Websites
List your sites on the portal’s Website page and paste the Grafana Faro snippet into their pages. Page loads, web vitals and JavaScript errors then arrive from your visitors’ browsers, from those sites only, with no password in the page.
Retention
How long we keep each signal is your choice.
Every signal starts with a default. On the portal’s Data policy page you set your own, per signal, up to the maximum below. Your plan may cap it lower.
Metrics
- Default
- 30 days
- Maximum
- 90 days
Logs
- Default
- 14 days
- Maximum
- 30 days
Traces
- Default
- 7 days
- Maximum
- 14 days
Drop what you do not need
Name the metrics and labels to drop before they are stored, and they are never stored. They still cross the network, so drop them in Alloy as well to save that too.
Shorter is final
Shortening a retention deletes the older data at the next compaction, and nothing brings it back. A change takes effect within about two minutes.
Multi-tenancy
One shared stack, a tenant for each customer.
Mimir, Loki and Tempo are shared, and each customer is a tenant of its own in them. What keeps tenants apart is enforced by our gateway, never left to the sender.
Stored under your tenant ID
Your data is written and read under your tenant’s ID. Other tenants can neither read it nor write into it.
The gateway sets the tenant
A request to your address gets through only with one of your logins, and our gateway overwrites the tenant header with your ID, whatever the sender wrote.
Your Grafana sees only your data
Your Grafana reads through the same gateway with your tenant’s login, so even its admins see your data alone. It cannot reach another customer’s Grafana or anything private in the cluster.
Only what you need is reachable
Writing, querying and your own alerting are routed to the stores. Their admin endpoints are not.
Limits per tenant
Each tenant has caps of its own, by default 10,000 metric samples, 4 MB of logs and 5 MB of traces a second, and a cap on active series. A tenant over a cap has only its own writes refused.
Tested on a live cluster
Our isolation test checks, on a running cluster, that one tenant can neither read nor write another’s metrics, logs and traces, and that the stores answer only through the gateway.
Where it runs
Stored in Germany, reached through Amazon CloudFront.
Where your telemetry is kept, and the providers on its way there.
Stored in Nuremberg
The cluster runs on Hetzner servers in Nuremberg, Germany, and Mimir, Loki and Tempo keep your data in Hetzner Object Storage in the same location.
Through Amazon CloudFront
Your agents, browsers and Grafana sessions connect to Amazon CloudFront over HTTPS. CloudFront ends that TLS connection at its edge location, opens a new HTTPS connection to our cluster and passes the request on. It caches nothing, and plain HTTP is refused.
HTTPS only through CloudFront
The cluster’s firewall accepts HTTPS only from CloudFront and from our admins’ own addresses.
No usage reports to Grafana Labs
Mimir, Loki, Tempo, your Grafana and the Alloys we run have their usage reporting switched off, and our CI fails if one is switched back on.
Run by a German company
Telopi is run by Optiop GmbH. We process data under German law and the GDPR.
See your own telemetry in Telopi.
Book a 30-minute demo, or look around the live one first.