My Always-On AI Agent System: Telegram, Ollama, and an Obsidian Vault on a Mac Studio
I built a 6-agent AI system that runs 24/7 on my Mac Studio. Telegram for input, Ollama for inference, Obsidian for memory. Here’s the full architecture — ho...
I Let Claude Code Handle Everything I Was Too Scared to Touch
This looks risky. This looks like it’s only for engineers. That’s exactly what I thought — and exactly what AI is solving right now.
Read Article →Recent posts
5 Models Tested, 2 Deleted: What Actually Works for Local AI Agents on M2 Max
Leaderboard scores don’t tell you which models work for AI agents. I tested 5 local models on my M2 Max for real agent tasks — orchestration, coding, researc...
I Run 3 Local Models and 1 Cloud API — Here’s How I Route Between Them
Theory says hybrid LLM routing saves money. I built a system that actually does it — 6 AI agents, 3 local models, 1 cloud API, running 24/7 on a Mac Studio. ...
Running Llama 3.3 70B Locally: Hardware Requirements and Complete Setup Guide
Llama 3.3 70B is the most capable open-source model you can run at home — but it demands serious hardware. Here’s exactly what you need, what to expect, and ...
Stop Sending Everything to GPT-4: A 5-Factor Framework for Local vs Cloud LLMs
Stop sending everything to GPT-4. Five factors decide whether a task should run locally or hit a cloud API — here’s the framework to make that call in 30 sec...
LLM Cost Optimization: How to Reduce Your API Bills from $2,000 to $400/Month
A 5-person dev team was spending $2,000/month on LLM APIs. After applying these 7 techniques, they cut it to $400 — without losing output quality. Here’s exa...
Hybrid LLM Architecture: Save 50-70% on AI Costs with Smart Routing
Most teams route every AI task to GPT-4 or Claude. That’s like hiring a senior engineer to do data entry. Here’s the hybrid architecture that cuts API bills ...