In open-source agentic coding harnesses, there seems to be 2 top choices: Pi and OpenCode. Yesterday I tested Laguna XS 2.1 (Laguna-XS-2.1-Q4_K_M.gguf) in llama-cpp on my local system in OpenCode. Today I tested Pi instead.
They both worked. Overall I felt that Pi seemed faster and more responsive to my prompts. Both harnesses seemed to handle most tool calls correctly. And on my system Laguna XS was fairly fast and capable in refactoring Rust programs with both harnesses.
Some differences:
• Pi seemed to be more thorough in checking its work (reading in the file and verifying changes were made correctly).
• I liked how Pi's interface updated faster for terminal commands (you can see each token being generated).
So which one is better? There isn't a simple answer to this—they both work, and they are both open source. I tend to give the edge to Pi for now; and given that Pi does not try to sell any AI subscriptions to its users, I prefer it.