BreezeRun

Hey Product Hunt! 👋

I am glad to share BreezeRun (MVP) which I recently built.

A few weeks ago, I was using Claude.ai to help me prototype some algorithms. It wrote the code, Claude.ai ran it directly in the web interface, showed me the output as images directly, and debugged with real world data directly with the browser — all in one seamless flow. I thought, "This is incredible." Then I realized the limitation. When I needed to analyze proprietary datasets or work with customer code under NDA, cloud-hosted AI simply wasn't an option. Compliance requirements meant the data couldn't leave our infrastructure — full stop. I needed something that could run entirely on-premise, with the flexibility to use locally-hosted models when security demanded it.

So I started building what I wished existed. An AI assistant that could do everything the Claude.ai do — write code, execute it, iterate on errors — but running entirely on my own machine. I wanted to choose my LLM provider (or run models locally with Ollama). I wanted code execution sandboxed in Docker so I could trust it with real tasks. And I wanted to see exactly what the agent was thinking at each step.

BreezeRun is the result. It's still early — definitely a hobby project with rough edges — but it works. The ReAct agent can break down complex problems, write code across 13 language environments, recover from errors, and show you everything it's doing in real-time.

If you're a developer who values owning your tools and data, I'd love to hear what you think. And if this resonates with you, contributions are very welcome.

GitHub: https://github.com/zysoong/breez...