Replay storage costs

VoiceGateway’s conversation replay captures every STT chunk, LLM token, TTS frame, and conversation-state snapshot for a session, when capture is turned on. See CLI: replay for how to enable capture (attach(snapshots=True), LiveKit only, default off) and inspect a captured session. This page surfaces the on-disk storage cost so the trade-off between fidelity and footprint is visible before you set per-project retention.

What gets stored

Four tables (src/voicegateway/models/replay_event_model.py), one row per captured event. Each table serializes its event fields into a single payload TEXT column rather than separate typed columns:

Table	`payload` JSON keys	Typical payload size
`replay_stt_events`	`text`, `is_final`, `alternatives`	100-400 bytes
`replay_llm_tokens`	`token_text`, `role`, `is_tool_invoke`, `tool_args_partial`	80-200 bytes
`replay_tts_frames`	`frame_duration_ms`, `underrun`, `voice_id`	80-120 bytes
`replay_state_snapshots`	`system_prompt`, `message_history`, `tool_call_in_flight`, `structured_output_collected`	500-5000 bytes (depends on prompt and history size)

replay_stt_events, replay_llm_tokens, and replay_tts_frames also carry id, session_id, t_ms, provider, cost_usd, created_at, tenant_id. replay_state_snapshots is narrower: no provider or cost_usd column. Column overhead runs roughly 80-120 bytes/row; the index on (session_id, t_ms) that each table carries adds an estimated 30-50% on top of payload size.

Per-minute estimate

For a typical voice conversation (caller speaks for half the time, agent for half, normal-cadence LLM with a 500-token system prompt):

Modality	Events per minute	Bytes per minute
STT (partials + finals)	30-60	8 KB-24 KB
LLM tokens	200-500	30 KB-80 KB
TTS frames (20-50ms each)	600-1500	60 KB-180 KB
State snapshots (1/sec cap)	60	30 KB-300 KB

Total: roughly 130 KB-580 KB per minute of conversation. The floor applies to short, crisp exchanges; the ceiling applies to chatty agents with long conversation histories. The design target of 30-100 KB/min is achievable at the floor; realistic agents will land closer to the ceiling. If you find yourself trending above 500 KB/min consistently, the per-project replay.enabled: false toggle is the fastest mitigation.

Worked example

A solo developer running 100 voice calls per day, averaging 5 minutes each:

100 calls/day x 5 min/call x 200 KB/min = 100,000 KB/day ~ 100 MB/day

At the default 90-day retention:

100 MB/day x 90 days ~ 9 GB total replay storage

At AWS S3 standard storage prices (~$0.023/GB-month):

9 GB x $0.023 ~ $0.21/month

On the local SQLite database (no cloud markup), the cost is the disk byte cost: ~

0.01/GB-month on a developer SSD, so ~

0.09/month for the same 9 GB. A team agency running 10,000 conversations per day at 3 minutes average scales linearly: ~3 TB at 90-day retention, ~

70/month on S3 standard or ~

30/month on local disk. At that point the retention_days knob matters: dropping to 30 days cuts storage to one-third.

Tuning knobs

Three per-project knobs in voicegw.yaml’s replay: block influence storage:

Knob	Default	Effect
`enabled`	`true`	Set `false` to skip capture entirely for this project.
`retention_days`	`90`	Age replay rows out after this window. Lower to reduce footprint linearly.
`flush_size_events`	`500`	Batched writes. Smaller flushes more often; larger holds more memory. No effect on long-term storage volume.
`buffer_size_events`	`5000`	Per-session in-memory cap. Above this limit, the oldest buffered events are dropped and a warning is logged server-side every 100 drops. Nothing in the dashboard currently surfaces the drop count.

The enabled toggle is the binary on/off. The retention_days knob is the gradient lever. The buffer_size_events and flush_size_events knobs trade off memory pressure and write batching but do not change long-term storage volume.

Dashboard storage view

GET /api/replay/storage returns per-project replay byte totals. The dashboard surfaces this as a breakdown so you see the cost in real time:

{
  "total_replay_size_bytes": 9234567890,
  "by_project": [
    {"project": "acme", "replay_size_bytes": 8000000000},
    {"project": "default", "replay_size_bytes": 1234567890}
  ]
}

Storage

The full SQLite schema including the replay tables.

CLI: replay

Inspect a specific session with voicegw replay.

Cost reconciliation

Verify recorded costs against provider invoices.

Architecture index

Overview of all architecture pages.

API

CLI

MCP server

Architecture

Help

Contributing

What gets stored

Per-minute estimate

Worked example

Tuning knobs

Dashboard storage view

Storage

CLI: replay

Cost reconciliation

Architecture index

​What gets stored

​Per-minute estimate

​Worked example

​Tuning knobs

​Dashboard storage view

​Related pages

Storage

CLI: replay

Cost reconciliation

Architecture index

What gets stored

Per-minute estimate

Worked example

Tuning knobs

Dashboard storage view

Related pages