Skip to main content

Federated Telemetry

If you already run Prometheus, Loki, and Jaeger (or equivalents), you don't have to ship raw telemetry to InfraSage at all. Federated mode inverts the data flow: your stack keeps custody of the raw data, and InfraSage pulls what it needs.

This works because InfraSage's detection core is aggregates-only by design. The detectors, the dashboards, and the watchdog all read one-minute aggregates, never the raw firehose. Raw logs and traces are needed only as evidence during an investigation, and evidence can be fetched at investigation time from wherever it lives.

How it works

ConcernIngest modeFederated mode
Metrics for detectionAggregated from your OTLP/Prometheus pushPulled from your Prometheus on an interval, written straight to aggregates
Raw logs (RCA evidence, Explore)24h transit buffer in InfraSageQueried live from your Loki, results cached ephemerally, citations kept
Raw traces24h transit bufferQueried live from your Jaeger/Tempo
Raw data custodyInfraSage (24h TTL)Yours, entirely
Evidence retention24hWhatever your stack retains

The RCA's search_logs and get_traces tools route through per-tenant evidence adapters. The analysis doesn't know or care where the evidence came from, and every adapter failure is loud and attributed, never silently empty.

Configuring sources

Console → Connect → Federated sources, or the API:

GET/POST/PUT/DELETE /api/v1/federated/sources

A source declares a kind (prometheus, loki, jaeger), an endpoint URL reachable from InfraSage, optional auth, and a poll interval (metrics sources only). The capability matrix on the Connect page shows exactly which platform features each source kind powers.

:::warning Reachability is on you InfraSage queries your endpoints from its egress IPs, so your network has to allow it. Private stacks typically expose a mTLS-guarded or IP-allowlisted endpoint. The pull fails loudly per query when unreachable, and the source's status reflects it. :::

:::note Rollout status Federated mode ships behind FEDERATED_TELEMETRY_ENABLED and is currently enabled per deployment during partner onboarding. Talk to us before flipping it in production. :::

When to choose which

Pick ingest mode if you want one vendor, minimal moving parts, and same-day evidence without running your own log store. It's the better fit for teams without an existing stack.

Pick federated mode if you already have an observability investment, a data-custody requirement, or retention needs beyond 24h raw. Your stack's retention becomes InfraSage's evidence window, and interactive exploration load stays on infrastructure already sized for it.

The two compose: some services pushing OTLP, others federated, in the same tenant.