Blog

What we are building, what broke along the way, and what we changed our minds about.

Run CLI Agents In Parallel With Git Worktrees

Two agents sharing one working tree is a stash fight waiting to happen. Here is the model we use instead.

The case for building around the CLIs you already use instead of wrapping one provider's chat in an Electron shell.

Closing a window should never kill a three-hour agent run. The trick is who owns the PTY, and it is not the UI.

Tabs full of agents are useful until they need to coordinate. How orchestration protocol v2 adds ownership, gates, and a real coordinator.

Hitting a provider limit mid-run is the worst way to learn your quota. We put Claude, Codex, Kimi, Grok, Cursor, and more in the status bar.

We did not want to invent another agent protocol. If it runs in a terminal, it runs in Alera - and the big ones get first-class treatment.

Five agents in parallel and the fans spin up. The Resource Manager names the culprit per tab - including the time it blamed us, 26 times over.

Every new worktree needs the same ritual: copy the .env, install dependencies, bootstrap. We made the repo describe it once.

Review belongs next to the work. How the Pull Requests panel keeps GitHub, GitLab, and Azure DevOps scoped to the worktree that produced the branch.

A chatty agent can stream output for an hour straight. That is a frame budget problem, and on Linux it is worse than you think.

The mobile companion is honest about what it is: a window into your runtime host, not a phone IDE. Here is what it does today.

Why we bet on Flutter, Rust, and Ghostty instead of shipping another Electron shell around CLI agents.

Why we started building a native workbench for CLI coding agents, and what you can do with it today.