I Built an Open-Source AI Agent That Benchmarks Itself (And It's Actually Good)

No API costs. No VC funding. Just 3,000 lines of Python and a llama.cpp backend.

Hello I've build yet another terminal IDE too that works out-of-the-box with a local LLM backed.

It's called open-agent. Running Qwen 3.6 35B on 6GB VRAM with pretty incredible results for a privacy centered setup/environment.

Here's the deep dive into Why's and How's that'll hopefully help some fellow builders.

The Problem With Every Agent Framework