Arize Phoenix

What it is

Arize Phoenix is an open-source LLM observability and evaluation platform: trace visualization, prompt/agent quality evals, and dataset curation, built on OpenTelemetry's GenAI semantic conventions. It ingests traces over the standard OTLP traces endpoint (/v1/traces). Nothing Phoenix-specific is required on the sending side.

Consume traces from Ingate

Create an export config pointing at your Phoenix collector's OTLP traces endpoint:

curl -X POST https://api.ingateai.com/api/v1/exports \
  -H "X-Ingate-Key: sk-ingate-..." \
  -H "Content-Type: application/json" \
  -d '{
    "name": "phoenix",
    "kind": "otlp",
    "endpoint": "http://<phoenix-host>/v1/traces",
    "include_sources": ["proxy", "api"]
  }'

kind: "otlp" batches trace entries and POSTs them as application/x-protobuf, which is exactly what Phoenix's OTLP receiver expects. Spans carry the OTel GenAI semantic convention attributes (gen_ai.provider.name, gen_ai.request.model, gen_ai.usage.input_tokens/output_tokens, gen_ai.input.messages, gen_ai.output.messages, gen_ai.user.id, gen_ai.session.id) that Phoenix reads natively to render prompt/completion pairs, token usage, and latency in its trace UI, no attribute mapping needed on either side.

include_sources controls which gateway traffic feeds Phoenix: proxy (passthrough/translation requests), api (agent-monitoring/SDK ingest), otel (spans Ingate itself received via its OTel receiver). Omit the field to default to ["proxy", "api"].

Reference

• Trace Exports: how gateway trace streaming works
• Exports API: full config shape, include_sources semantics, hot-reload behavior