openclaw-workspace

Personal OpenClaw workspace for Brian (台台小助手). A Taiwanese AI assistant configured for Discord interaction, voice transcription, YouTube summarization, and local AI workflows.

Summary

This workspace powers a personal AI assistant running on a home server with NVIDIA A2000 8GB GPU. The assistant communicates via Discord, processes voice messages with local Whisper, summarizes YouTube videos, and manages media files on a NAS.

Key capabilities include:

• Real-time Discord voice channel conversations
• Local speech-to-text with GPU acceleration
• YouTube video download and transcript summarization
• Local OCR and text-to-speech
• NAS storage management with automated cleanup

Included Assets

Core Configuration

• AGENTS.md — Workspace startup and session management
• MEMORY.md — Long-term memory and user preferences
• SOUL.md — Assistant identity and personality
• TOOLS.md — Local tool configuration notes

Skills

• skills/discord-voice/ — Discord voice channel plugin (real-time voice conversations)
• skills/video-transcript-downloader/ — YouTube transcript fetching

Memory Logs

• memory/2026-03-28.md through memory/2026-04-07.md — Daily session notes

How to Use

This workspace is managed by OpenClaw. On session startup, the assistant automatically reads:

1. SOUL.md — identity
2. USER.md — user context (if exists)
3. memory/YYYY-MM-DD.md — recent daily notes
4. MEMORY.md — long-term memory (main session only)

Key Paths

Notes

• GPU Priority: OCR, video transcoding, STT, TTS, and speech understanding use local A2000 GPU first
• Taiwanese Identity: Assistant identifies as Taiwanese, prefers zh-TW voice (HsiaoChenNeural)
• File Sharing: HTTP servers use 24-hour expiry
• Storage Cleanup: /mnt/nas_data/tmp/ auto-clears files older than 30 days (3:00 AM daily)