Composable architecture
Use the desktop tray, command line, bridge daemon, host-native integration, or a combination. Speech remains a reusable service instead of one chat app feature.
Build a private Linux voice workflow
A voice assistant that does not own your assistant.
AverVOX is the reusable speech layer around the model, application, or agent you already prefer. Keep voice local, swap LLM endpoints freely, and use the same hotkeys throughout Linux.
Local speech processingOpenAI-compatible LLMsDesktop + CLINo vendor lock-in
What it gives you
Own the speech layer, choose the intelligence layer.
AverVOX separates microphone and speaker workflows from the model service, giving you a stable interface while the AI stack changes.
Local by default
Raw microphone audio, STT, TTS, voice activity detection, and text insertion stay on the machine. You choose where conversation text is sent.
Local, LAN, or remote models
Run the LLM on the same laptop, a more capable workstation, a private server, or a provider—without replacing the Linux speech workflow.
Setup
Assemble the stack without hand-gluing every process.
Start with a supported Linux desktop, then connect the speech layer to the model endpoint that matches your privacy and performance goals.
Choose where inference runs
Use Ollama, LM Studio, vLLM, LocalAI, another OpenAI-compatible server, or a remote inference provider.
Install the local speech layer
AverVOX supplies Faster Whisper, Piper, VAD, hotkeys, desktop insertion, streaming TTS, and the conversation state machine.
Add automation when useful
Use the bridge CLI and daemon for scripts, or install the Hermes and OpenClaw packages when an agent should reuse the same speech engine.
Use the right machine for each job
Speech and inference do not have to share a GPU—or even a computer.
AverVOX can run responsive local speech on the Linux desktop while a different LAN machine handles a larger model.
Desktop
Hotkeys, microphone, Faster Whisper, text insertion, and TTS stay close to the user.
Network
Only conversation text and streaming reply text cross to the endpoint you configured.
Inference host
A larger workstation or server can run the model without becoming your everyday desktop.
Common questions
What to know before you start.
Is AverVOX itself the LLM?
No. AverVOX is the speech and desktop interaction layer. You connect it to an LLM endpoint or command-line workflow of your choice.
Can everything remain offline?
Yes, when the model endpoint is also local. Speech processing already runs locally; point Converse at an on-device or LAN model server for a fully local path.
Why not wire Whisper, an LLM, and Piper together manually?
You can. AverVOX packages the recurring production concerns around that pipeline: desktop hotkeys, VAD, text insertion, streaming sentences, interruption, profiles, configuration, warm model serving, capability discovery, and host integrations.
Can I use it without Converse?
Yes. System-wide dictation, selected-text TTS, and CLI speech work independently of any LLM.
How does Pro change the assistant experience?
Pro adds Kokoro, wake-word activation, persistent sessions, per-profile system prompts, a LAN server, a dashboard, conversation transcripts, and a daily dictation log.
Start with the workflow
Build the assistant around your tools—not the other way around.
Use the free core as the stable speech layer. Upgrade only when the Pro automation, persistence, and voice-quality features earn a place in your daily workflow.