Destinations

Exactly two destinations exist. Pick with OBSERVABILITY_TYPE.

	postgres	http
Where traces go	A PostgreSQL table, telemetry.proxy_traces	Any HTTP endpoint, one POST per batch
Readable through Axium's API	Yes, GET /v1/observability/traces	No (409 on that route)
Schema managed by Axium	Yes, created and migrated at startup	No, the receiver owns it
Formats	Table rows	envelope (JSON) or ndjson
Computed fields (latency_ms, tokens_total, cost)	Generated columns	You compute them
Authentication to the destination	In the DSN	OBSERVABILITY_HTTP_HEADERS
Timeout per write	10 s to acquire a connection	5 s per request
Good for	Getting started, small and medium volumes, querying from the console or SQL	Feeding ClickHouse, BigQuery, Datadog, Splunk, S3, a SIEM, or a collector

Common behaviour (batching, drops, no retries) is described in the overview. Delivery details for each:

• PostgreSQL destination
• HTTP destination

Not available

There is no console, file, OpenTelemetry/OTLP, Langfuse, OpenInference or vendor-native exporter in this version. All of those are reachable through the HTTP destination with a collector in between: an OpenTelemetry Collector or Vector can receive the NDJSON stream over HTTP and fan it out.

Choosing

• Start with postgres on the same database. It costs nothing extra, the console can show traces, and telemetry.proxy_traces is a normal table you can query, export or replicate.
• Move to http when you want traces in the same place as the rest of your telemetry, when volume grows beyond what you want in the transactional database, or when you need durable delivery through a collector.
• You cannot run both at once. To get both, use http into a collector that writes to PostgreSQL and elsewhere.