Evaluate performance feasibility of current setup

[uggy-ug]

While testing the current setup I’ve noticed that even very simple requests - like "describe this repo" or "create a markdown file" take noticeably long to complete.

I’d like to open a discussion on:

• Whether these delays are expected given the current architecture
• If there are optimizations we can make to improve response times

Cost is something that makes me curious about use cases for this type of application.

[rishu685]

Right now, Open SWE feels slow even for small tasks because it’s doing a lot behind the scenes — not just asking the AI once, but breaking your request into steps, spinning up a sandbox, reading files, sometimes running code, and then pulling it all together.

We can make it faster by:
• Skipping repeated work (cache repo info)
• Running steps in parallel
• Using smaller/faster models for easy parts
• Keeping sandboxes “warm” instead of starting from scratch every time

As for cost — it comes from both AI token usage and the time the sandbox runs. It’s worth it for bigger jobs (like code changes or deep repo summaries) but overkill for quick, one-line questions.