Here was the problem I encountered: in my chat app, many users worked with files, and at first I relied on OpenAI’s built-in code interpreter. But it started causing issues, especially around file generation. Around the same time, new tools were released—like Claude Code and OpenAI Codex—that handled a wide range of tasks much more effectively, but they depended on shell-based execution. So, I instead created an internal tool which would use those models provided with virtual shell access. Shellify provides options to either execute commands directly or use adapters like Claude Code to work on files within a sandboxed shell environment.
Here is the link: https://shellifyai.com/