Conductor runs the agents. Aura reads what they did.
Conductor is a Mac app that launches several Claude Code or Codex agents at once, each in its own Git worktree, with one place to watch them and review the diffs. Aura runs agents the same way and then does the part Conductor leaves to you: it reads each change at the logic level, checks it against the goal the agent was given, and records a verdict you can hand to someone else.
What each one is for
What Conductor is great at
- Worktree-per-workspace
The isolation model is the right one, and Conductor got there first. Every agent gets a real worktree, so two of them can move on the same repo without stepping on each other.
- A genuinely nice Mac app
Native, quick to start, and the workspace list is the clearest way anyone has drawn "here are my six agents and what each is doing".
- Linear and GitHub in the loop
Pull an issue in as context, open the PR from the app, hand a plan from one agent to another. The ticket-to-PR path is short.
What Aura adds on top
- A semantic diff, not a text one
Aura reads the change as functions and call edges. A rename is a rename, not 200 red lines and 198 green ones.
- A verdict on the goal
Each task carries the goal it was meant to deliver. Aura checks the commit against it and says proven, partial, or not wired — before the merge, not after the incident.
- A deletion guard
If an agent removes an exported function nothing asked it to remove, the commit stops and names the callers. That is the failure mode a diff review is worst at catching.
- Not only Mac, not only GitHub
macOS and Linux today, Windows in beta. Any Git remote, or none — a repo that never leaves your laptop works fine.
Dimension by dimension
Side by side
When to pick which
Pick Conductor if…
- You are on a Mac, on GitHub, and the workspace list is exactly the UI you want.
- Linear-issue-to-pull-request is the loop you run all day.
- You review diffs by eye and are happy doing that.
Pick Aura if…
- You have shipped an agent change that passed review and broke something anyway.
- You need to show someone — a reviewer, an auditor, a customer — why a change is correct.
- You are not on macOS, or not on GitHub.
- You want the tool you depend on to be open source.
Keep Conductor. Add the control layer.
They overlap enough that most people pick one. If you already like Conductor, the honest test is a week of running both on the same repo: same agents, same tasks, and see which one catches the change that should not have shipped.
Aura & Conductor
Is Aura a Conductor clone?
The worktree-per-agent idea is shared, and Conductor deserves credit for popularising it. What Aura adds is the layer after the agent stops: an AST-level read of the change, a check against the goal, and a signed record. That is the part Conductor does not do.
Does Aura work on Linux or Windows?
Yes. macOS (Apple Silicon and Intel) and Linux (arm64 and x86_64) ship today; Windows is in cert-free beta. Conductor is macOS only.
Do I need GitHub to use Aura?
No. Aura works against any Git remote, or against a repository that never leaves your machine.
Is Conductor free?
It has been free so far and runs on your existing model subscriptions. Aura is free in public beta, and its engine, CLI and editor extension are Apache-2.0, so the licence does not change under you.
Ship with proof, not hope
Free public beta for macOS and Linux. Point it at a repo and keep driving Conductor, now with a semantic diff, a verdict and a record on every change.