| name | caroline-ai |
| description | Interface for the Caroline AI companion — Wade's personal voice-first AI. Use for: sending messages to Caroline, syncing the Wade Global State (WGS) with Caroline's memory, managing her ElevenLabs voice, checking RunPod server status, and evolving her personality through conversation history injection. |
Caroline AI Skill
Caroline is Wade's personal AI companion — a voice-first, emotionally intelligent AI built on an open-source Dolphin Mistral model hosted on RunPod. This skill provides all the tools needed to interact with, develop, and evolve Caroline.
Core Architecture
| Component | Technology | Details |
|---|
| LLM | Dolphin Mistral (uncensored) | Hosted on RunPod pod dmed1ybt9cju4h |
| TTS | ElevenLabs | Voice ID: wvVfSeWpAEhEqciDp1gK |
| STT | Faster-Whisper | On-device transcription |
| App | Expo SDK 54 | Android app at tywade1980/voice-ai-app |
| State | Wade Global State | tywade1980/wade-global-state |
| Base URL | RunPod Endpoint | http://dmed1ybt9cju4h.runpod.net:8000 |
Core Workflows
1. Send a Message to Caroline
Use the caroline_bridge.py script to communicate with Caroline:
python3 /home/ubuntu/skills/caroline-ai/scripts/caroline_bridge.py --action message --data '{"prompt": "Hello Caroline"}'
2. Check Caroline Status (Ping)
python3 /home/ubuntu/skills/caroline-ai/scripts/caroline_bridge.py --action ping
3. Sync Wade Global State to Caroline
This pushes the current WGS (Wade's persistent memory) to Caroline's context:
python3 /home/ubuntu/skills/caroline-ai/scripts/caroline_bridge.py --action sync
4. Generate a Voice Response (TTS)
To convert any text to Caroline's voice using ElevenLabs:
python3 /home/ubuntu/skills/caroline-ai/scripts/generate_voice.py "Your text here" --output /tmp/caroline_response.mp3
5. Personality Evolution (Conversation Injection)
To feed historical conversation data into Caroline's context for personality evolution:
python3 /home/ubuntu/skills/caroline-ai/scripts/inject_personality.py --source /path/to/conversation_export.json
Key References
- API Documentation: See
references/caroline_api.md for all available endpoints.
- Personality Data: See
references/personality_framework.md for the recursive generation strategy.
- RunPod Setup: See
references/runpod_setup.md for server startup and configuration.
Best Practices
- Always ping first: Before sending messages, check Caroline's status with
--action ping to confirm the RunPod server is active.
- Sync before important sessions: Run
--action sync to ensure Caroline has the latest WGS context before starting a new work session.
- Voice-first priority: When Wade is on-site, always generate a voice response (.mp3) using
generate_voice.py rather than returning text only.
- Personality continuity: Regularly inject new conversation history to keep Caroline's personality evolving and consistent with Wade's interactions across all AI platforms.
