Toscan — AI Chief of Staff
Running 24/7Toscan is my autonomous AI chief of staff — a system that runs 24/7 on dedicated infrastructure, orchestrating multiple AI agents to get real work done.
Built on OpenClaw with a 3-tier model system: Claude Opus for architecture and complex reasoning, Claude Sonnet for coding and research, and Claude Haiku for housekeeping tasks. All running on a Claude Max subscription at zero per-token cost.
Toscan manages an entire orchestra of specialized workflows: Deep Research (5 parallel workers + citation verification), Code Shipping (architect → coders → PR management), Content Pipeline (research → writing → image generation → publishing), and Security Audits (parallel repository scanning).
In a single afternoon, Toscan orchestrated 13 sub-agents simultaneously — shipping code, writing articles, generating images, and running security audits. Total extra cost: $0.
Tech Stack
Key Features
- ✓3-tier AI model orchestration (Opus/Sonnet/Haiku)
- ✓13+ concurrent sub-agents in parallel
- ✓Deep Research Orchestra (46+ sources, citation-verified)
- ✓Code Shipping Orchestra (feature branches, PRs, reviews)
- ✓Content Pipeline Orchestra (research → write → images → publish)
- ✓Proactive monitoring (email, calendar, heartbeat checks)
- ✓Voice calls via Retell AI + Twilio
- ✓Multi-channel communication (Telegram, Discord)
Challenges
Building reliable autonomy required careful guardrails: agents work on feature branches (never push to main), all PRs require human approval, and a memory system maintains context across sessions. The compound effect is real — each new orchestra makes the whole system more productive.