Mehmet Özgen ÖZDOĞAN·5d agoKovan: A Local Framework That Combines vLLM, Other LLM Engines, Rules, and Durable User Memory…Most local AI setups start as “one model behind one endpoint.” That works until real people use it. Someone asks the same thing every day…
Mehmet Özgen ÖZDOĞAN·5d agoLocal AI This Month (November 2026): New Models, New Optimization Tricks, and What’s Happening…Every few weeks I do a sweep of the local AI space to see what actually changed versus what just got a new press release. This is the…A response icon2A response icon2
Mehmet Özgen ÖZDOĞAN·Nov 10Harness Engineering for Local AI: The Part Around the Model That Decides Whether It WorksHere’s a pattern I keep running into. Two people run the same local model on the same GPU. One gets an agent that finishes tasks, fixes its…
Mehmet Özgen ÖZDOĞAN·Nov 8Building an exo-Style Distributed LLM Cluster on Ubuntu Server VMs With Mixed GPUs: Architecture…You have a few Ubuntu Server VMs. Each one has a different GPU attached through passthrough: maybe a 24 GB card in one, a 16 GB card in…
Mehmet Özgen ÖZDOĞAN·Nov 8Local AI Is Getting Stronger: Distributed Inference, Agent Concurrency, and the Projects Worth a…A year ago, “local AI” mostly meant one model on one laptop, great for privacy and tinkering, awkward for anything with real concurrency…A response icon1A response icon1
Mehmet Özgen ÖZDOĞAN·Oct 28AI-Powered Attacks Are Here: What Businesses and Governments Should Actually Do About ThemFor years, “AI will supercharge cyberattacks” was a conference-keynote prediction. It’s now a line item in incident reports. In the last…A response icon1A response icon1
Mehmet Özgen ÖZDOĞAN·Oct 28When Your AI Agent Phones Home: Threats Cloud-Connected Agents Pose to Local SystemsMost of this series has argued for running AI locally. This post makes the other side of that argument concrete. Plenty of people will keep…
Mehmet Özgen ÖZDOĞAN·Oct 17AgentOps for Local AI: Making Your Agents Observable, Traceable, and DebuggableThe first time one of my multi-agent runs went wrong, I had nothing to look at. The pipeline had taken forty minutes, the final report was…
Mehmet Özgen ÖZDOĞAN·Oct 14The Missing Layer in AI Coding Agents: Requirements, Traceability, and VerificationMost coding agent guides start at the same place: you type a task, the agent writes code, you skim the diff, maybe run it. That works for a…A response icon1A response icon1
Mehmet Özgen ÖZDOĞAN·Oct 13Building a Private AI Network Operations Center: Local LLMs for Topology, Firewall Logs, IaC, and…Network data is some of the most sensitive data a company owns. Topology maps, VLAN layouts, routing tables, firewall rules, and raw…