#voice enablement

1 messages · Page 1 of 1 (latest)

shy vault
#

two quick questions about integrating openclaw-voice (brokemac79's repo) with an OpenClaw instance running in Docker on Windows 11, currently on v2026.4.26:

What is the correct OPENCLAW_URL to use for the voice server? Specifically — what HTTP endpoint and port does OpenClaw expose for agent chat turns? (e.g. http://localhost:3000/api/chat or something else?)
The openclaw-voice README mentions a 403 missing scope: operator.read error when using a /v1/ endpoint, noted against v2026.3.28 and suggesting a CLI fallback. Is this fixed in v2026.4.26, or is the CLI fallback still the recommended path?

sinful coralBOT
#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

stuck arrowBOT
#

Short version: use the Gateway OpenAI-compatible endpoint, not /api/chat.

# voice server running on Windows host:
OPENCLAW_URL=http://127.0.0.1:18789/v1/chat/completions

# voice server running in Docker on Windows:
OPENCLAW_URL=http://host.docker.internal:18789/v1/chat/completions

OPENCLAW_MODEL=openclaw/default
OPENCLAW_AUTH_BEARER=<your OpenClaw gateway token or password>
OPENCLAW_CLI_FALLBACK_ENABLED=false

OpenClaw’s Gateway default port is 18789, and /v1/chat/completions runs on that same Gateway HTTP/WS port. The endpoint is disabled by default, so enable:

#
{
  gateway: {
    http: {
      endpoints: {
        chatCompletions: { enabled: true },
      },
    },
  },
}

Then restart/doctor the gateway as needed. Current docs also say bearer auth with gateway token/password restores full operator scopes, including operator.read/operator.write, so on 2026.4.26 the CLI fallback should not be the default path anymore. Use fallback only if you actually still get the 403 scope error. (docs.openclaw.ai)

Smoke test: