Local AI and n8n setup
Mac Mini AI stack that survives reboots.
A diagnostic for Ollama, n8n, AnythingLLM, ChromaDB, Telegram bots, launchd persistence, and written handoff on macOS.
Setup risk map
| Area | What I check before touching the machine |
|---|---|
| Persistence | Whether n8n, ChromaDB, AnythingLLM, and the Telegram workflow should run through Docker Compose, LaunchAgents, or LaunchDaemons. |
| Ollama on M-series | Model choice, memory footprint, GPU use, context length, and whether workloads interfere with the desktop session. |
| Network | LAN-only access, webhook exposure, tunnel choice, firewall rules, and Telegram bot callback path. |
| Data storage | Where ChromaDB, n8n data, model files, backups, and logs live so they are recoverable. |
| Validation | One end-to-end test: Telegram message -> n8n -> local model/RAG -> response -> log. |
$29 diagnostic deliverable
- Mac service plan: Docker Compose vs
launchd, with restart and login-session notes. - Folder and backup layout for n8n, ChromaDB, AnythingLLM, and Ollama models.
- Security checklist: FileVault, firewall, local admin user, webhook exposure, and secrets handling.
- Fixed implementation estimate for the full $500-style setup.