v0.9.2 published · controlled-testing prerelease

VoxRouter turns your devices into a private voice network for agents.

Run one local-first hub on Windows, macOS, Docker, or source. Pair Android phones and Wear OS watches over trusted LAN HTTPS, route speech to OpenClaw/Hermes or OpenAI-compatible models, then speak back through the voice provider chain you control.

Windows installer · Android + Wear OS clients · free Edge voice default
setup + live turn router online
0.9.2
Published v0.9.2 controlled-test artifacts
Windows installer · Android APK · Wear APK · Apple Silicon DMG
SETUP migrate OpenClaw config, then discover agents auto
PAIR scoped phone + watch credentials tls
HEALTH errors, 429s, quota headers stats
streamed audio reply chunk 1 · 90ms
Current release signal

v0.9.2 is published with its Windows smoke evidence recorded.

Automatic OpenClaw config migration OpenRouter sign-in without hunting for a key Cold-agent validation up to 120 seconds 28 keyscan tests in CI
wake loop router whisper agent chain tts chain audio
Abstract blue, cyan, amber, and mint signal paths branching from a central VoxRouter hub on a dark technical grid.
Signal path One voice turn, routed visibly from microphone to agent to spoken reply.
Why it is different

A voice stack with local control at every boundary.

The latest release removes a fragile OpenClaw upgrade step, makes first-run setup easier to understand, and lets operators authorize OpenRouter from the hub without copying an API key.

new migration path

Recommended setup repairs a moved OpenClaw config, then finds the agent.

When OpenClaw 2026.7.1-2 moves mcpServers to mcp.servers, VoxRouter creates a backup, migrates supported entries idempotently, then validates a protected gateway for up to 120 seconds so a cold local agent is not misreported as missing.

Provider stats expose real routing behavior.

The console and /v1/providers/stats show attempts, errors, 429s, quota headers, and fallback rescues since router start.

Every device pairs without copying the admin key.

Android, Wear OS, and the native iPhone client use hub trust confirmation, local-CA HTTPS, and separately revocable device credentials. Apple Watch transport and push-to-talk remain explicitly hardware-gated roadmap work.

Explore

Everything between the microphone and the reply.

Five deep-dives, one router. Compose all of it per agent in a single YAML file.

Quick start

Start with the installer, or drop to Docker/source.

For most users, the published v0.9.2 Windows installer is the shortest path: choose recommended setup, read the new VoxRouter orientation, let it migrate a compatible OpenClaw config, wait for a cold protected agent when needed, and test from the localhost console. OpenRouter users can sign in from Settings with a one-time PKCE flow; operators can still run the router from Docker or source.

~/voxrouter
> Run VoxRouter-Setup-v0.9.2.exe
> Choose Recommended automatic setup
> Keep "Find and connect my OpenClaw agents now" selected
✓ localhost admin: 8088 · paired clients: 8443
# v0.9.2 also migrates config and waits for cold agents
# source/docker path still works
$ cp .env.example .env
$ python scripts/generate_api_key.py --env .env
$ docker compose up --build
# curl a turn
$ curl -s localhost:8088/v1/voice/text \
-H "authorization: Bearer $VOICE_PROXY_API_KEY" \
-d '{"agent":"dante","text":"Say hello","speak":true}'
{ "chat_backend": "a2a", "tts_provider": "edge" }

Your voice stack, your rules, your hardware.

Clone it, drop in a config, and route voice to any agent you run.