> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.thymia.ai/sentinel/llms.txt.
> For full documentation content, see https://docs.thymia.ai/sentinel/llms-full.txt.
# Sentinel
Sentinel is thymia's real-time monitoring SDK for voice-based AI agents. It
analyses audio and transcripts as they stream, runs safety policies, and
returns actionable results — all without storing recordings.
Sentinel is currently in early access. For access and configuration,
contact [support@thymia.ai](mailto:support@thymia.ai).
## How it works
Sentinel sits alongside your voice agent pipeline. You stream user and agent
audio (plus transcripts) over a WebSocket connection. Sentinel runs
biomarker models and safety policies in real time and returns results through
callbacks.
## Key capabilities
* **Real-time biomarker extraction** — Helios, Apollo, and Psyche models run
on streamed audio to measure distress, emotion, and clinical indicators.
* **Safety policies** — configurable rules that trigger when biomarker
thresholds are crossed, returning classification levels and recommended
actions.
* **Progress updates** — periodic callbacks showing how much speech has been
captured and when biomarker results will fire.
* **Platform agnostic** — works with any voice pipeline. Drop-in integrations
for VAPI, Gemini Live, and LiveKit.
## Quickstart
Pick your voice platform to get started:
### VAPI
Telephony-first voice agent platform.
### Gemini Live
Google Gemini Live API with 24 kHz audio.
### LiveKit
Open-source WebRTC infrastructure.
## Next steps
* [Policies](/sentinel/concepts/policies) — understand how safety
policies work.
* [Sentinel Docs](https://thymia-ai.github.io/thymia-sentinel-integrations/1.1.0/) — full SDK reference and integration guide.
***
