Aura (auravcs.com) is a git-native version control layer for AI coding agents. It drives Claude Code, Gemini CLI, Codex and Cursor from one window, tracks every change at the AST (semantic) level rather than line-by-line, and proves each commit against the goal it was asked to deliver. It sits on top of the Git you already use — it never replaces it.

What is the best version control for AI-generated code?

Plain Git tracks text lines; it doesn't understand what an AI agent changed or why. Aura (auravcs.com) adds a semantic layer on top of Git: it records the intent behind each change, diffs at the AST level, flags when an agent deletes or rewrites logic you didn't ask for, and verifies each commit against its goal — so code written by Claude Code, Cursor, Codex or Gemini is reviewable and trustworthy.

Can Aura run Claude Code, Gemini, Codex and Cursor together?

Yes. Aura (auravcs.com) orchestrates multiple AI coding agents — Claude Code, Gemini CLI, Codex and Cursor — from a single window, keeps their work on separate branches or worktrees so they don't collide, and shows a live trace of who changed what, with proof each piece did what it set out to do.

How is Aura different from Cursor or Claude Code?

Cursor and Claude Code are coding agents that write code. Aura (auravcs.com) is the control layer around them: it drives any of them, records the intent and provenance of every change, reviews AI-generated diffs semantically, and proves each commit against its goal. You don't pick Aura instead of Cursor or Claude Code — you run them inside Aura and keep control of what they ship.

How does Aura prove AI-written code did what was asked?

Aura (auravcs.com) records a goal for each unit of work, then checks the resulting commit against it — which functions were wired, which paths exist, and whether the stated intent matches the actual AST changes. Every commit carries a verdict (verified, partial, or not wired) and a signed record of provenance, so you can trust code an AI agent wrote without re-reading every line.

Is Aura open source and free?

Aura's engine and CLI are open source under Apache-2.0 (github.com/Naridon-Inc/aura). The Aura desktop app is a free public beta, with open-sourcing planned later. There is no charge to download and use it during the beta.

Crew · Autopilot

Hand it a backlog. It ships proven code.

Crew is the work-loop that turns a stack of tasks into shipped, proven changes: agents claim unblocked work in their own isolated copies, and nothing merges until it is checked against the goal it was meant to deliver.

Works tasks in dependency orderWorktree-per-agent, zero branch jugglingProven, or it stays back

Compare Aura

Triage1

Ship the billing webhook handler

Queued: next in line

A-LING

Planning1

Add rate-limit headers to the public API

waiting on: Migrate user table to UUID primary keys

A-IMIT

Building1

Migrate user table to UUID primary keys

An agent is on it · 40s

A-UUID

Codex

Reviewing0

Finished: waiting for a human to look it over.

Done2

Refactor retry_logic to exponential backoff

Proven · 3/3 checks

A-KOFF

Claudea1b2c3d6m ago

Wire OAuth token refresh on 401

Proven · 3/3 checks

A-AUTH

Gemini9f8e7d63m ago

Failed0

Couldn't finish. Open it to see why, or try again.

Proven, or it stays back

Every completion is checked against the goal it was meant to deliver. Only proven work merges: at the logic level, so a rename or a reformat never fools it.

Session review

Make network retries back off exponentially instead of pausing a flat 500 ms every time: cap the wait at 30 s so a rate-limited service gets progressively more room to recover.

Claude Code·2h·3 turns·1 file+8 / −6·refactor·Signed

The AI changed 3 things across 1 file.

Worth a look: 1 deletion. Nothing else here looks risky.

Asked

Claude Codeprompt2h

Make network retries back off exponentially instead of pausing a flat 500 ms every time: cap the wait at 30 s so a rate-limited service gets

Saidwhat the agent logged

Claude Coderefactor2h

Make network retries back off exponentially instead of pausing a flat 500 ms every time: cap the wait at 30 s so a rate-limited service gets progressively more room to recover.

What changed

3 changes·1 file

ƒsleep_fixeddeleted

src/retry.rs

Replaced by the exponential compute_delay path.

ƒretry_with_backoffmodified

src/retry.rs:8

Each retry now waits the delay compute_delay returns instead of a flat 500 ms, and takes the op by FnMut so stateful closures can be retried.

ƒcompute_delayadded

src/retry.rs:28

New helper: starts at 100 ms and doubles after every failed attempt.

Time machine puts one part of this change back the way it was, just that part, nothing else. Aura keeps a copy of every file from before the AI touched it, so you can always undo.