Why we built it this way
Why Ember works the way it does, written by the people who build it and use it every day: the decisions, the mistakes and what we changed.
Why we built Ember: from Cursor to Claude Code to our own editor
We were deeply invested in Cursor, moved to Claude Code, and found the VS Code extension wasn't enough for the way we work. This is what we kept, what we missed and what we built.
One task, one branch, one pull request: how Ember organizes parallel agents
Why Ember's agents never commit on your base branch, why background work gets a git worktree and no window, and what we changed after the first design failed its own review.
Cleanup never takes your work: a rule we wrote after losing a file
An automatic cleanup in Ember once removed a worktree that still held an uncommitted edit and an untracked file. This is the rule we wrote the same day and how Ember checks it.
Your agents, on your phone: why Ember's mobile app talks to your own computer
How Ember's phone app is built: a hub on your own machine, a one-time pairing code, per-device tokens, and the same next step the desktop shows. And what we deliberately keep off the phone.
Your Claude subscription, not an API key: how Ember handles sign-in and permissions
Ember runs on your Claude Pro or Max plan through the local Claude Code CLI. It never implements sign-in, never reads credentials, and strips API keys from the agent's environment. Here is why, and how approvals work.