Launch HN: Superset (YC P26) – IDE for the agents era
GitHub avipeltz ● Covered by 2 sources
Three YC founders built Superset, an open-source IDE for running multiple AI coding agents at once. It fixes the messy git-worktree juggling devs were already doing by hand.
Avi, Kiet, and Satya kept running into the same wall: once CLI coding agents like Claude Code and Codex got good enough to trust with real work, they wanted to run several at once, one triaging GitHub issues, another building UI, a third reviewing PRs. The problem wasn't spinning up the agents. It was that they, and apparently half their engineer friends, had all independently hacked together similar scripts around git worktrees just to keep the isolated copies of a repo from turning into chaos.
Superset started a few months back as a fairly modest open-source terminal for managing those worktrees. It's since grown into something bigger: an IDE built specifically for orchestrating agent work across multiple worktrees, repos, and now machines. The founders say the real lesson from months of user feedback wasn't that people needed more agents running simultaneously. It was that nobody could keep track of what five or ten agents were actually doing, which ports they'd claimed, which diffs were ready, which PRs needed a human's eyes. So Superset added task and issue tracking that follows work through the full loop, from issue to agent to diff to PR to review, without anyone losing the thread.
The newer piece is Remote Workspaces, still in beta, which lets agents run on a separate machine instead of eating your laptop's RAM and CPU while you still control everything from the Superset desktop app. To make that work, the team rebuilt the guts of their Electron app as a headless Hono server, so the same interface can talk to whatever client you're on, desktop, web, eventually mobile, without duplicating logic everywhere.
That mobile piece is coming next, alongside a beefed-up CLI and smoother remote workflows. The pitch, basically, is that if you're not physically at your dev machine you should still be able to check in on your agents, look at their progress, and nudge them in a new direction from your phone. It's a fairly clear bet that agentic coding is becoming something you manage rather than something you sit and watch.
My take
The interesting tell here isn't the product, it's that a handful of engineers independently reinvented the same worktree scripts before anyone built this properly, which says the tooling gap around parallel agents was real and overdue. Betting on remote workspaces and mobile control assumes people will trust agents enough to let them run unsupervised on someone else's machine, which is a bigger leap than the IDE itself, and worth watching closely as review fatigue becomes the actual bottleneck instead of compute.
Read more about this at: GitHub
