OTLP endpoint

Radicas ingestion is a standard OTLP endpoint behind an authenticating gateway. Anything that speaks OTLP can send to it — the Radicas SDK, a raw OpenTelemetry exporter, or your own Collector.

Endpoint and protocols

The hosted endpoint is https://ingest.radicas.io. It accepts OTLP over HTTP (standard signal paths: /v1/traces, /v1/metrics, /v1/logs) and OTLP over gRPC. Whether gRPC is served on port 443 of the same host or on a dedicated host/port is to be confirmed — this section will be updated before it matters; the SDK and the OTLP/HTTP examples below work today as shown.

Authentication

Every request must carry your ingest key as a bearer token: Authorization: Bearer followed by the key. With generic OpenTelemetry exporter configuration:

export OTEL_EXPORTER_OTLP_ENDPOINT="https://ingest.radicas.io"
export OTEL_EXPORTER_OTLP_HEADERS="authorization=Bearer ___INGEST_KEY___"

Sign in to see your real key pre-filled in this snippet. Requests without a valid key are rejected at the gateway. Key issuance, reveal, rotation, and revocation are covered in API keys.

Headers that matter

• Authorization — the only credential the gateway looks at. Tenant identity is resolved from this key, server-side.
• Content-Type — for OTLP/HTTP, application/x-protobuf (binary protobuf, the default for most exporters) or application/json.
• Tenant headers — any client-supplied tenant header is stripped at the gateway and replaced with a trusted value derived from the key. You cannot (and never need to) set tenant identity yourself; see tenancy and ingestion.

Keep gen_ai attributes intact

If anything transforms your telemetry en route — a Collector pipeline, a proxy, a sampling processor — do not strip these; they drive cost computation and per-feature attribution:

• gen_ai.usage.* — token counts, the input to estimated cost
• gen_ai.request.model — which price-book entry applies
• span parent/child links — the agent span tree
• radicas.feature — per-feature allocation

Batching, retrying, and relaying are all fine; see bring your own Collector for the supported patterns.

Limits

Per-tenant rate limiting is enforced at the gateway. The exact limits, maximum payload sizes, and recommended retry/backoff behaviour are to be confirmed and will be documented here — they are not published yet, and any numbers you may find elsewhere should not be relied on.