Best For
BNI chapter Vice Presidents or officers with JavaScript/CLI technical skills, running a macOS machine that stays on 24/7, and willing to manage multiple external service accounts. Particularly suited for LINE ecosystem users in Taiwan/Asia-Pacific.
How I Actually Use It
Primarily as an architecture reference. The raw/wiki layered knowledge base design (immutable raw data + LLM-continuously-rewritten structured pages) is a reusable knowledge management pattern. The SHA256 Skill Pinning concept applies to supply chain security. The dual-brain cost control architecture (free tier for high-frequency low-value tasks, paid model for high-quality long text) is worth studying.
Where It Is Strong
- Dual-brain architecture: GPT-5.4 (Codex OAuth free tier) handles real-time chat while Claude handles high-quality knowledge compilation, keeping costs minimal
- The raw/wiki layered design separates append-only raw data from an LLM-rewritten structured wiki layer with internal links, enabling semantic queries without embeddings
- Five-layer fuzzy matching handles Chinese name format variations, typos, nicknames, and proxy detection. This solves a real pain point in BNI attendance tracking.
- Full BNI workflow automation: Zoom join → attendance → transcription → report → stats → LINE notification, end-to-end
- SHA256 Skill Pinning prevents framework updates from breaking core functionality, showing solid supply chain security awareness
Where It Fails
- Single-machine architecture: Mac shutdown = service down, no high availability
- Uses Computer Use to simulate personal LINE account operations, which risks account bans
- macOS-only: LaunchAgent + pdftotext + Computer Use all tied to macOS
- Extremely early stage: 6 commits / 8 stars / single maintainer / no automated tests
- Codex OAuth free tier is not officially guaranteed and may be revoked
- Designed for one chapter only, no multi-organization coordination
Pricing, Difficulty, and Risk
- $6-17 USD/month (Recall.ai transcription + LLM API usage)
- Difficulty is high: requires setting up 7+ external services (OpenAI, Claude CLI, Telegram Bot, LINE, Recall.ai, Google Sheets, Apify)
- Risk factors: LINE account ban from simulated operations, single machine with no HA, single maintainer may abandon, PII stored locally without encryption
Verdict
A proof-of-concept BNI administrative automation solution. The architecture (dual-brain + raw/wiki + SHA256 pinning) is worth studying and borrowing from, but the project is too immature for production use. Best suited as an architecture reference or customization starting point for technically capable users.
Source
- GitHub: https://github.com/AlexChanshuo/BNI-VP-Autopilot
- License: Not specified