Cost Tracking
Arkenos tracks the cost of every API call made during voice sessions — STT minutes, LLM tokens, and TTS characters.How Costs Are Tracked
- The agent worker logs usage events in real-time via
POST /api/usage/events - Each event records the provider, service type, units consumed, and calculated cost
- The backend aggregates costs for dashboards and reports
Provider Rates
Rates are defined inbackend/app/cost_rates.py and can be updated in one place.
| Service | Provider | Unit | Rate | Hourly Equivalent |
|---|---|---|---|---|
| STT | Deepgram | per minute | $0.0077 | $0.46/hr |
| STT | AssemblyAI | per minute | $0.0078 | $0.47/hr |
| STT | ElevenLabs | per minute | $0.0167 | $1.00/hr |
| LLM | Google Gemini 2.5 Flash | per 1K tokens (blended) | $0.00086 | ~$0.86/1M tokens |
| TTS | Resemble AI | per character | $0.00004 | $40/1M chars |
The Gemini rate is a blended estimate assuming a ~3:1 input-to-output token ratio. Actual rates: 2.50/1M output tokens.
Usage Event Structure
Dashboard Views
Cost Summary
Dashboard → Costs shows:- Total cost across all sessions
- This month’s cost
- Breakdown by provider (pie chart)
Cost Timeline
View costs over time with daily, weekly, or monthly granularity.Cost by Agent
See which agents cost the most, helping you optimize prompts and configurations.Per-Session Breakdown
Dashboard → Call Logs → [Session] shows the cost breakdown for a single call:| Provider | Service | Units | Cost |
|---|---|---|---|
| AssemblyAI | STT | 2.5 minutes | $0.0195 |
| Google Gemini | LLM | 1,250 tokens | $0.0011 |
| Resemble AI | TTS | 450 characters | $0.0180 |
| Total | $0.0386 |
What’s Not Tracked
Arkenos currently does not track costs for:- LiveKit — connection minutes (~$0.0005/min per participant). These are billed by LiveKit directly.
- Twilio — per-minute telephony charges for inbound/outbound calls. These are billed by Twilio directly.