Question: why does readme recommend using openai-completions api for coding agent usage.

[njbrake]

I noticed in the README.md that the default recommendation is to use the openai-completions api within opencode and pi.

There is no motivation listed and I would like to know if there was a reason this was chosen?

From my experience working on https://github.com/mozilla-ai/any-llm I have seen that the OpenAI Completions API is generally poorly suited for agentic use cases and that the Anthropic Messages API or the newer OpenResponses (open source spec of OpenAI Responses API) are the recommended ways to use LLMs for agentic purposes.

pi.dev supports the use of the Anthropic Messages API and based on my current understanding, it would be best for ds4 to recommend this as the setup for using this codebase within pi.dev.

Any thoughts or disagreements before I post up a Pull Request for the change? Thanks!

[fry69]

FYI: There is some related discussion in the Responses API PR #91

[antirez]

The reason for now is purely the fact that is the most tested code path. The completion API has no technical merits, and more work will happen in that space soon. For now I wanted to stabilize the inference as much as possible.