Hi HN, my co-founder and I run Akkento, a small self-funded team building Hyperlane, an independent IDE built on the VS Code source (Code OSS), made for merging worktree-based agent development with actual IDE features.

Agent orchestrators give you a worktree based workflow where several agents run in parallel, but they are not IDEs (they are not meant to be), so the moment an agent finishes you have to leave and review the diff somewhere else, actually debug the code and go through a process to sanity check it. In a commercial team, that review step isn't optional which is the problem we came across, where you have 2 editors open at the same time for the same job. IDEs are where you actually read, debug and profile the code, but none of them give you the worktree workflow. We are hoping to fix that by merging the two together in an IDE with tooling for profiling, building and testing.

Try it out: https://hyperlaneide.com/

Hyperlane runs on macOS (Apple Silicon), Windows and Linux (x64 and arm64, .deb and .rpm packages). It's free to download, no account needed to run it, and you bring your own models.

In Hyperlane, every worktree gets its own tabs, editors, windows and terminals, so you can context switch quickly and each worktree is isolated as a project of its own. We also built inter worktree caching so N worktrees don't cost N copies of the repo on your drive.

We're building this as a full IDE and added a Node and React profiler, full project indexing for codebase search, codebase intelligence, formatting via Prettier, error lens for showing inline errors, and native vite test support. Long term we want the whole IDE surface to be as good as the worktree side, and there's plenty still missing.

While building Hyperlane, we kept a running list of features missing from our daily editors, or scattered across several of them. We added a strong git client, a three way conflict editor, per worktree blame and diff, and pull requests inside the editor, with support for GitHub, GitLab, Codeberg and self hosted Forgejo or enterprise hosts. We're not vendor-locked to anything and we use multiple different forges ourselves. The same goes for agents. Hyperlane works with Claude Code, Codex, opencode, and any CLI or ACP agent. There is also a design mode for making live visual edits to a localhost site and handing those changes to an agent as context.

As we are a fork of VSCode I know this puts us in the "another fork of vscode" category. Originally we spent months writing our own editor from scratch, with our own LSP and semantic highlighting. But after months of work, we realised we couldn't live without extensions. So we built VS Code extension support into it (literally), only to face a choice to either finish the editor, or finish and maintain the extension system which was a huge undertaking on its own. So we threw that work away and rebuilt on Code OSS to focus on the features we actually wanted in the editor, which let us spend our time on the worktree and agent side of things. Happy to argue about the tradeoff, it was not a fun call to make, especially on the side of the performance loss.

It's very early and there are rough edges, but we are hoping to build Hyperlane together. I'll be looking over this thread, and you can also raise tickets at (https://github.com/Akkento/hyperlane) and I want to note that Hyperlane is closed source, and the repo is releases and issue tracking only. There is no source in it, so I don't want anyone going looking and feeling misled. Thank you for reading this long text :)