Engine modes
Last modified: 25 July 2026
Three modes
| Mode | Where speech-to-text runs | When to pick it |
|---|---|---|
sidecar | A processor over HTTP — your Docker container, or the Sniffer Studio cloud | Default choice. Keeps inference off the game loop and scales independently. |
embedded | Inside the plugin, in your server JVM | One server, no Docker, spare CPU. |
none | Nowhere — capture is disabled | The shipped default, and a clean off switch. |
Cloud subscriptions are always sidecar: transcription runs on the Sniffer Studio
fleet. The processor token is the entitlement; cloud mode does not need a licence key. Self-host
and embedded modes require the licence key.
Embedded
engine:
mode: embedded
embedded:
tier: standard
profile: safe
| Setting | Values | Meaning |
|---|---|---|
tier | lite · standard · multilingual | lite is the least CPU and memory; standard is the English default; multilingual is heavier and ships the English and Czech rule packs. |
profile | safe · balanced · fast | Thread budget for inference: 1, 2 or 4 threads. Start at safe and only raise it if the server has spare cores. |
The model downloads on first start. A failed download or a hash mismatch leaves embedded moderation unavailable rather than half-working, and the log says so. Tier, profile and language list are startup-only: change them and restart the server.
Sidecar
engine:
mode: sidecar
processor:
url: http://127.0.0.1:8090/v1/moderate
token: 'your-processor-token'
language: auto
url— your processor's/v1/moderateendpoint. The default is a container on the same host. On cloud this is assigned by your subscription and the generated file carries a comment instead of a value.token— the bearer token for this server, from the processor'stokens.json. It is what scopes a request to your licence; without it the processor answers 401.language—autoevaluates every installed rule pack (English and Czech today), or pinen/csif you know what the server speaks. A pinned language outside your plan is refused by the backend.
Running the container yourself: Docker processor.