Feature request

Problem:
messages.tts.openai accepts apiKey, model, and voice, but does not support a baseUrl (or baseURL) field to override the OpenAI TTS endpoint.

This prevents using OpenAI-compatible local TTS servers (e.g., Kokoro, Piper, or other self-hosted TTS behind an OpenAI-compatible API wrapper) through OpenClaw's built-in tts tool.

The ElevenLabs TTS config already supports baseUrl — the OpenAI TTS section does not.

Proposed solution:
Add an optional baseUrl field to messages.tts.openai:

{
  messages: {
    tts: {
      provider: "openai",
      openai: {
        apiKey: "local",
        baseUrl: "http://127.0.0.1:8190/v1",  // <-- new field
        model: "kokoro",
        voice: "onyx"
      }
    }
  }
}

Use case:
Running a local FastAPI server that implements POST /v1/audio/speech (matching OpenAI's TTS API) to serve locally-hosted TTS models (Kokoro, F5-TTS voice cloning, etc.) at zero cost and zero latency.

Current workaround:
Calling the local TTS server via exec/curl and sending the audio file manually, bypassing the built-in tts tool entirely.

Environment:

• OpenClaw 2026.3.2
• Config validation rejects baseUrl under messages.tts.openai as an unrecognized key