Track 1 - Episode 3 - Email to Meeting Notes with LmStudio, AMD Strix Halo, N8N, Budibase
In Episode 3 of our 2026 On-Prem Series, we’re moving from infrastructure to implementation. It's one thing to have a server running; it's another to have it solving business problems like summarizing
There series is noted here https://www.newsdailyai.studio/youtube
In Episode 3 of our 2026 On-Prem Series, we’re moving from infrastructure to implementation. It’s one thing to have a server running; it’s another to have it solving business problems like summarizing meeting notes from an email box and turning them into actionable tasks.
In this video, we cover:
Agent Optimization: How to fix slow n8n agents by tweaking tracing callbacks and switching to faster models like OSS 20B.
The “Skills” Architecture: Moving from simple prompts to a “Company Skills” database in Supabase for consistent AI performance.
BuddyBase UI: Building a front-end for your intranet so non-technical staff can edit AI prompts and manage project tasks.
Docker Networking Deep Dive: Solving the 307 Redirect issue by correctly using internal Docker hostnames instead of external IPs/Tailscale addresses.
LM Studio Tweaks: Adjusting context length and KV cache to prevent “mid-generation” crashes.
Hardware & Software Stack:
Hardware: AMD Ryzen “Strix Halo” (optimizing for A-Series APUs).
Networking: Tailscale for secure remote access to the intranet.
Backend: Supabase (Postgres & Artifacts).
Automation: n8n 2.x using MCP (Model Context Protocol) for task management.
Join the Community: This is part of a year-long track for 2026. Members get access to the private Discord, shared n8n workflows, and BuddyBase templates used in these videos.
🏷️ Tags #OnPremAI #n8n #BuddyBase #Supabase #SelfHosted #LocalLLM #LMStudio #DockerNetworking #AIAutomation #2026Tech #PrivateAI #MCP #ModelContext Protocol #AMDStrixHalo
Timestamps
0:00 - Introduction & The “Rough Cut” Philosophy 1:00 - Recap: Private LLMs, Coolify, and SSL 1:50 - Exploring Alaris: A Potential Cloud-Managed Pivot? 3:42 - Addressing n8n Speed Issues & Local Processing Hurdles 4:32 - Goal: Email Automation Foundation 6:45 - Using Tailscale for Secure Local Access 8:50 - n8n Optimization: Fixing Tracing & Callback Latency 11:15 - Model Selection: Why OSS 20B for Agent Routing 13:50 - Transitioning from “Prompts” to “Company Skills” 19:00 - n8n 2.0: Managing Artifacts & MCP Tools 23:45 - The Task MCP: Separation of Concerns for AI Tools 27:10 - Optimizing LM Studio: Context Length & Temperature 32:00 - Troubleshooting: Why Bob Didn’t Get His Email 40:50 - Debugging the 307 Redirect (Docker Networking vs. Tailscale) 54:00 - Visualizing Internal Docker Networks 58:20 - Setting Up BuddyBase in Coolify 1:01:20 - Building the Skills Editor in BuddyBase 1:06:30 - Ejecting Blocks & Customizing Long Form Fields 1:09:50 - Final Thoughts: Sim AI vs. n8n for Knowledge Bases