mahimailabs/voicegateway — explained in plain English
Analysis updated 2026-05-18
Track exactly how much each voice AI call costs across speech-to-text, LLM, and text-to-speech providers.
Reconcile your recorded usage numbers against real provider invoices to catch billing surprises.
Monitor voice call quality metrics like latency, interruptions, and dead air across a live pipeline.
Bill different teams or clients separately using per-tenant cost attribution and scoped API keys.
| mahimailabs/voicegateway | a-bissell/unleash-lite | abhiinnovates/whatsapp-hr-assistant | |
|---|---|---|---|
| Stars | 1 | 1 | 1 |
| Language | Python | Python | Python |
| Setup difficulty | moderate | hard | hard |
| Complexity | 3/5 | 4/5 | 3/5 |
| Audience | developer | researcher | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires a LiveKit Agents project already running before you swap in the import.
VoiceGateway is a self-hosted observability and cost tracking layer for voice AI applications built on LiveKit Agents, a framework for building real-time voice bots. It acts as a drop-in replacement for one import, so you change a single line of code and immediately gain detailed visibility into what every voice call is costing you and how it is performing. Voice AI costs are unusually hard to track because three separate services each bill differently: speech-to-text charges by audio seconds, large language models charge by tokens, and text-to-speech charges by characters. VoiceGateway captures all three together and breaks every call down to a per-modality cost in cents. A reconcile command lets you compare your recorded numbers against actual provider invoices. Beyond cost, the dashboard tracks voice-specific quality metrics that text applications do not have: latency measurements across the speech-to-text-to-LLM-to-speech pipeline, interruption rates, dead air time, and talk-over. You can replay any past conversation and scrub through each chunk with timing and cost attached. For teams managing multiple clients or projects, it supports per-tenant cost attribution and virtual API keys scoped to specific teams. It also includes guardrails that can detect sensitive personal information or prompt injection attempts in real time during speech recognition. The whole system runs on your own infrastructure with no data leaving your stack. A local dashboard opens at port 9090. It supports 11 providers across cloud and local options for each modality. The project is released under the MIT license, so it can be used, modified, and reused freely, including for commercial purposes.
A self-hosted dashboard that tracks the real cost and quality of every voice AI call across speech-to-text, LLM, and text-to-speech.
Mainly Python. The stack also includes Python, LiveKit Agents.
MIT license: free to use, modify, and reuse for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.