Problem Description

Currently, TTS (Text-to-Speech) configuration in OpenClaw is global and shared across all agents. The messages.tts configuration in openclaw.json applies to all agents and channels.

Current Limitations

1. Global TTS Settings: All agents use the same TTS provider, voice, and settings
2. No Per-Agent Voice Selection: Cannot assign different voices to different agents
3. No Voice ID Parameter Support: Even when using ElevenLabs with baseUrl pointing to a custom endpoint, there is no way to pass different voice_id values per agent

Use Case

We have multiple agents (e.g., main agent, coder agent) and want each to have a distinct voice/tts configuration:

• Main agent: Female voice (warm, friendly)
• Coder agent: Male voice (professional, clear)

Proposed Solutions

Option 1: Per-Agent TTS Config

Add TTS configuration support in agents section:

{
  agents: {
    main: {
      messages: {
        tts: {
          provider: "elevenlabs",
          elevenlabs: {
            voiceId: "female_voice_id"
          }
        }
      }
    },
    coder: {
      messages: {
        tts: {
          provider: "elevenlabs", 
          elevenlabs: {
            voiceId: "male_voice_id"
          }
        }
      }
    }
  }
}

Option 2: Agent ID Parameter

Allow passing an agentId parameter in the TTS request to select the appropriate voice from the TTS service.

Option 3: Voice ID Override via Directive

Enhance the existing [[tts:voiceId=xxx]] directive to work with custom endpoints via baseUrl.

Expected Behavior

Each agent should be able to have its own:

• TTS provider (elevenlabs/openai/edge)
• Voice ID / voice settings
• Other TTS-related configurations

Environment

• OpenClaw version: 2026.3.2
• Channel: Telegram
• TTS Provider: Custom ElevenLabs-compatible endpoint (ChatTTS)

Additional Context

This feature would be particularly useful for:

1. Distinguishing between different agents in group chats
2. Matching agent personality to voice characteristics
3. Providing accessibility options for users who prefer specific voices

Generated by OpenClaw

• Instance: zdeMacBook-Air.local (macOS 26.3, arm64)
• Gateway: ws://127.0.0.1:18789
• Agent: main (MiniMax-M2.5-highspeed)
• Channel: Telegram
• Date: 2026-03-07