pi-ollama

pi-ollama connects the pi coding agent to Ollama’s native chat API. It handles streamed responses and tool calls so a local model can participate in the same loop: receive context, request a tool, read its result and continue.

The work includes the less visible parts of that connection, such as incomplete responses, retry behavior and context handling. Those details affect whether an integration stays useful beyond a simple chat.

The provider handles Ollama’s response format directly and translates it into the events pi expects. That puts streaming and tool-call behavior in one place where it can be inspected and adjusted.

Running a model locally does not make every model equally capable of agent work. The integration provides the connection; the model’s ability to choose tools and supply sensible arguments remains something to test.

The extension has picked up users who have reported issues and submitted PRs. I’ve enjoyed working through those with them, and I’d like the project to become something people feel welcome to contribute to.

When a contribution needs more work, I try to explain what needs attention and give the contributor a chance to improve it. Taking the time to work through a change together matters to me. I want people to see their work through and have a reason to come back.

That is the kind of community I’m hoping to build around the project. It is still early, but the conversations and contributions have been a rewarding part of maintaining it.

I built pi-ollama with coding-agent assistance and continue to develop and maintain it. I work through user reports, review changes and help contributors get their work ready to merge. The ongoing work matters to me as much as getting the first version running.