Building a Local Agent Stack with Llama.cpp and Pi.dev
2026-08-08 (updated 2026-08-14) #agentic #llama.cpp #llm #pi.dev #self-hosted
In this post I am sharing some workflow tips and resources I gathered while working with local agents.
I spent a long time ignoring the agentic craze. Most of the early frameworks felt like over-engineered wrappers that either relied on intrusive cloud APIs or were too fragile for real development work. I did myself build an experimental agentic terminal app in the early days before even LLMs had tool calls, where I explored some of the ideas and wishes I envisioned in a theoretical future agent based IDE. However I was never going to use any of them connected to a cloud based service. I finally started paying attention when local models seemed to reach a tipping point in tool-call accuracy and reliability, specifically around the release of Qwen 3.5 35B.