Switching to Pi Agent Harness

A developer has switched from OpenCode to Pi Agent Harness, citing Pi's faster speed, minimalism, and pure open-source nature, and expressing concern that OpenCode is focused on selling AI subscriptions. The developer reports that the Laguna XS 2.1 coding model runs at over 40 tokens per second on a 12 GB Nvidia GPU system using llama-cpp, and recommends Pi and llama-cpp over subscription-focused tools like Ollama and OpenCode.

I have used both Pi and OpenCode in the past, and although I liked Pi I tended to just use OpenCode out of habit. OpenCode has become more efficient over time, and it works well enough. However, Pi seems to be even faster and more minimal, and also is a more pure open source project. One thing that concerns me about OpenCode is that it seems to be interested in selling AI subscriptions . If I wanted to subscribe to something like that, I could just use OpenAI or Anthropic. I want to focus more on local model usage through llama-cpp . I used Laguna XS 2.1 on Pi, and it was even faster than in OpenCode. If you want my opinion, Laguna XS 2.1 is the best coding model for a 12 GB Nvidia GPU system. It is 20.3 GB and runs at over 40 tokens per second in llama-cpp . Some projects, like Ollama and OpenCode, seem focused on selling AI subscriptions . This is fine, but if you are not interested in that, it is probably best to use other software like llama-cpp and Pi.