Document Eleven Labs voice agent configuration for self-hosting

[cruzanstx]

Summary

When self-hosting Happy, the voice feature doesn't work because the Eleven Labs Conversational AI agent IDs are hardcoded to the Happy team's account:

// RealtimeVoiceSession.tsx:29-30
agentId: __DEV__ ? 'agent_7801k2c0r5hjfraa1kdbytpvs6yt' : 'agent_6701k211syvvegba4kt7m68nxjmw',

Self-hosters need to create their own Eleven Labs agent but there's no documentation on how to configure it.

Request

Could you please either:

1. Document the system prompt used for the Eleven Labs agent so self-hosters can create their own, or
2. Provide guidance on what configuration is needed (system prompt, tools, settings), or
3. Share the system prompt in the repo (e.g., docs/voice-agent-setup.md)

What I Found

From the codebase, I can see the agent needs:

Client Tools:

• messageClaudeCode - takes { message: string } to send messages to Claude Code
• processPermissionRequest - takes { decision: "allow" | "deny" } for permission handling

Dynamic Variables passed at session start:

• sessionId
• initialConversationContext (full session history)

Context updates sent during session:

• Session online/offline status
• New messages from Claude Code
• Permission requests
• Ready events (when Claude finishes work)

But the actual system prompt that tells the agent how to behave (translate rambling speech into structured requests, when to use tools, etc.) isn't in the codebase.

Environment

• Self-hosted happy-server
• Self-hosted happy-cli
• Building the mobile app from source

Thanks for the great project! 🙏