Observability

VoiceGateway includes three built-in observability features that run as middleware around every provider call. All three are enabled by default and can be toggled independently.

Configuration

observability:
  latency_tracking: true
  cost_tracking: true
  request_logging: true

The three flags

latency_tracking

Default: true

When enabled, VoiceGateway measures time-to-first-byte (TTFB) and total latency for every provider call. Latency data is stored in SQLite and available through the dashboard, CLI (voicegw status), and HTTP API (/v1/metrics).

When disabled, provider instances are returned without the latency monitoring wrapper. This reduces overhead slightly but removes all latency visibility.

Related config:

latency:
  ttfb_warning_ms: 500.0
  percentiles: [50.0, 95.0, 99.0]

• ttfb_warning_ms -- a warning is logged when TTFB exceeds this threshold
• percentiles -- which percentiles to compute and report

cost_tracking

Default: true

When enabled, VoiceGateway estimates the cost of each provider call based on usage (tokens, characters, audio seconds) and records it in the SQLite database. Cost data powers the dashboard cost views, the voicegw costs CLI command, and per-project budget enforcement.

When disabled, no cost records are written. Budget enforcement (budget_action) will not trigger because there is no spend data to compare against.

::: warning Disabling cost_tracking also effectively disables budget enforcement for all projects, regardless of their budget_action setting. :::

Related config:

cost_tracking:
  enabled: true
  db_path: ~/.config/voicegateway/voicegw.db
  daily_budget_alert: 100.00

request_logging

Default: true

When enabled, VoiceGateway logs metadata about each provider call: timestamp, provider, model, modality, project, latency, and cost. Logs are stored in SQLite and visible in the dashboard request log view and through the voicegw logs CLI command.

When disabled, no request log entries are written. The dashboard log view will be empty.

Disabling all observability

To run VoiceGateway with zero overhead from observability middleware:

observability:
  latency_tracking: false
  cost_tracking: false
  request_logging: false

This is useful for benchmarking raw provider performance or in environments where you handle monitoring externally.

Checking current settings

voicegw status

The status output includes which observability features are enabled.

See: voicegw.yaml Reference, Projects, Environment Variables