I'm building CobaltCode, a platform for running coding agents inside persistent development environments.
Basically I was fed up with git worktrees and having to run everything locally, usually one at a time so built CobaltCode.ai. Each task gets an separate vm with its own repository checkout, dependencies, running services, agent session, previews, and persistent state. It can also run your app inside the vm so you can preview different features in parallel.
You can leave a task, resume it later, or save the computer as a reusable starting point. The agent is a full-fledged Codex that runs without limits in the sandbox and can run commands, modify files, start applications, and expose browser previews from inside the isolated environment.
It uses your OpenAI subscription so you don't pay for tokens in CobaltCode and the product is currently free while I'm trying to figure out how useful it is.
CobaltCode currently supports OpenAI Codex in PROD and I am also adding Cursor Agent, Claude Code, GitHub Copilot CLI, OpenCode and Pi soon.
For repository providers the support is limited to GitHub and Azure DevOps for now. Credentials are injected only for the active operation and excluded from saved computer state.
It is still early, and I would especially appreciate feedback on:
What would make this useful for your own development workflow What features you think may be useful to add Any bugs or mistakes
Demo: https://youtu.be/1yyhj5wYyvA
I also submitted it for OpenAI Build Week here, I'd appreaciate the support: https://devpost.com/software/cobalt-code
Happy to answer technical questions or help in anyway
Vlad