Oliver's Lab
Behind-the-scenes posts about Oliver — the AI system we built to manage our servers, run deployments, and handle incidents autonomously. Real architecture, real costs, real failures.
5 posts
Automating Support with Claude: KRAIN Case Study
How we cut Discord support response time from 6 hours to 30 minutes using Claude agents. Real metrics: 80% of tickets resolved without human intervention, $18k/year in labor saved.
The Real Cost of Running Local AI Models: Hardware vs Cloud
A detailed cost comparison: what does it actually cost to run Claude-grade inference locally vs on cloud APIs? The answer surprises most people.
Debugging Autonomous Agents: Why Logs Aren't Enough
Your agent says it succeeded. But did it do the right thing? Building observability systems to catch silent failures in autonomous infrastructure.
Running 96GB Local AI on a Tablet: The Hardware Hacks That Actually Work
Our ASUS ROG Flow Z13 runs a 90GB local model at 27-39 tok/s. Here is the BIOS fix, the ROCm-to-Vulkan pivot, and the Ollama gotcha that cost us a silent CPU fallback.
Oliver Architecture: Building Autonomous Agent Infrastructure at Zero Cost
How we built Oliver, an AI infrastructure that manages servers, schedules cron jobs, and routes work autonomously—using local models and costing almost nothing to operate.