Private local socket
The daemon uses a Unix socket inside a mode-0700 directory with a mode-0600 socket, limiting access to the current user.
Local speech for agents
One AverVOX installation. Every agent that needs a voice.
Replace cloud speech APIs with local STT and TTS. Published packages connect Hermes Agent and OpenClaw to the same bridge used by scripts and custom applications.
Host-native packagesAutomatic daemon fallbackSource-visibleLocal speech processing
Choose a host
Install through the package system your agent already understands.
Each integration uses the host’s documented provider or plugin interface rather than screen-scraping or UI automation.
| Host | Speech capabilities | Published package | Start here |
|---|---|---|---|
| Hermes Agent | TTS, STT, streaming, cancellation | avervox-hermes on PyPI | Hermes setup → |
| OpenClaw | TTS, streaming, cancellation | @avervox/openclaw-plugin on npm and ClawHub | OpenClaw setup → |
| Odysseus | Converse endpoint guide | Documentation | Read the repository → |
| Your application | TTS, STT, streaming PCM, capability discovery | avrvx CLI + Unix socket | Developer guide → |
Warm by design
Load the speech model once—not on every reply.
A one-shot CLI process is ideal for occasional automation. Agents that speak repeatedly can use the private daemon socket to remove repeated Python startup and model-load delay.
- Integrations try the warm daemon first
- Automatic fallback to one-shot avrvx processes
- No required network service
- Cancellation and capability discovery are part of the bridge
Reference performance
Measured through the real host interfaces.
These numbers describe one documented test machine and are intended to make the methodology inspectable—not to promise identical results on every system.
Hermes · warm0.32sFirst audio
OpenClaw · warm0.36sFirst audio
Hermes · spawn2.6sFirst audio
OpenClaw · spawn2.6sFirst audio
Trust is part of the integration
Know what runs inside your agent.
The packages are source-visible and designed to pass text and audio through protected local interfaces. Review and pin the package version that enters your agent process.
Safer text transport
Integrations prefer stdin or protected temporary files rather than placing sensitive generated text in the process list.
Auditable source
The integration code, tests, package manifests, compatibility notes, and benchmark procedure are available in the public repository.
Integration questions
Local speech, without a second cloud vendor.
Does the host need AverVOX Pro?
No. The bridge CLI, warm daemon, Piper TTS, Faster Whisper STT, streaming, and capability discovery are included in OSS. Pro is detected automatically and exposes its additional voice capabilities where supported.
What happens when the daemon is not running?
The published integrations fall back to launching the avrvx command for the request, so the integration still works with additional startup latency.
Can a host detect OSS versus Pro?
Yes. avrvx --capabilities returns structured JSON describing the edition, license state, version, engines, voices, STT model, daemon status, and feature map.
Does AverVOX send agent text to a speech API?
No. Speech synthesis is local. The host gives the reply text to AverVOX; AverVOX produces or streams the audio locally.
Where should I report integration problems?
Use the public integration repository for package issues and support@avervoxpro.com for licensed Pro support.
Start with the workflow
Give your agent a voice without adding a speech vendor.
Install the host package, connect it to the local AverVOX bridge, and keep the speech path on the Linux machine.