Metrics Reference

The Prometheus metrics OM1 exports for observability.

Overview

OM1 exposes a Prometheus metrics endpoint so you can monitor the AI pipeline in real time (ASR/LLM/TTS/VLM latencies, knowledge-base queries, response quality, and the internal HTTP proxy).

The metrics server listens on port 9090 at /metrics. For how to bring up the bundled Prometheus + Grafana stack and the pre-provisioned OM1 Latency Monitoring dashboard, see Quick Start → Prometheus and Grafana Monitoring. This page is the reference for the metrics themselves.

If you see Address already in use on port 9090, another process (often a local Prometheus) is using it — stop it or free the port.

Most latency metrics come in two forms: a histogram (_seconds, for quantiles/averages over time) and a gauge of the most recent value (_last_seconds, handy for a live "current latency" panel).

Pipeline latencies

Metric	Type	Description
om1_llm_latency_seconds / _last_seconds	histogram / gauge	Latency of LLM responses.
om1_vlm_latency_seconds / _last_seconds	histogram / gauge	Latency of VLM (vision-language model) responses.
om1_tts_latency_seconds / _last_seconds	histogram / gauge	Time from a TTS synthesis request to the first audio chunk.

ASR & VAD

Metric	Type	Description
om1_asr_latency_seconds / _last_seconds	histogram / gauge	Latency from locally-detected (VAD) end-of-speech to the final transcript. See VAD & TTS Interrupt.
om1_asr_speech_duration_seconds / _last_seconds	histogram / gauge	Duration of speech activity from speech-start to speech-end.
om1_asr_utterance_end_latency_seconds / _last_seconds	histogram / gauge	Latency from speech-activity start to end-of-utterance detection.
om1_asr_parallel_transcripts_total	counter	Transcripts seen by the parallel ASR sensor, labeled by provider model and first-wins outcome.

Knowledge base

Metric	Type	Description
om1_kb_query_latency_seconds / _last_seconds	histogram / gauge	Latency of a knowledge base query (embedding + search).
om1_kb_embed_latency_seconds / _last_seconds	histogram / gauge	Latency of the embedding step of a query.
om1_kb_queries_total	counter	Total knowledge base queries, by outcome.

Response quality (live quality scorer)

These are emitted only when the quality scorer is enabled — see Tracer & Quality Scorer.

Metric	Type	Description
om1_quality_live_active_score	gauge	Most recent turn's coherence score: coherent = 1, marginal = 0.5, incoherent = 0.
om1_quality_live_turns_scored	counter	Per-run count of turns that produced a coherence score.
om1_quality_live_coherence_count	counter	Per-run count of prompt/response pairs by coherence label (coherent / marginal / incoherent).
om1_quality_live_input_classification_count	counter	Per-run count of user inputs by classification (positive / marginal / negative / not_addressed).
om1_quality_live_language_count	counter	Per-run count of scored turns by detected spoken language.

Internal HTTP proxy

Timings for OM1's proxy to upstream AI services:

Metric	Type	Description
om1_http_proxy_total_seconds / _last_seconds	histogram / gauge	Total proxy time.
om1_http_proxy_parse_seconds / _last_seconds	histogram / gauge	Time the proxy spent parsing the request.
om1_http_upstream_total_seconds / _last_seconds	histogram / gauge	Total upstream service time.
om1_http_upstream_ttfb_seconds / _last_seconds	histogram / gauge	Upstream time-to-first-byte.

The authoritative list is defined in internal/metrics/metrics.go.