Skip to main content
Development previewFeatures and availability may change while we test.

Arivaran product portfolio

Tools for parallel agents and endpoint teams.

Choose Git, Twin, Aran, Reach, or Beam by the job you need to do.

Arivaran Git

Fast, safe git worktrees for parallel AI coding, with MOWI write-intent protection.

Open Arivaran Git
Linux · XFSmacOS · APFSBetaWindows · NTFSBeta

Plain git

Copied12.2s
git clone repo.git
git worktree add ../feature-a
git worktree add ../feature-b
$EDITOR ../feature-a/app.ts

Working tree ready

The agent blocks until it finishes.

Agent working
Copied up frontThe whole tree
Disk used810 MB

Every agent pays for a full copy before it can start.

Arivaran Git

Lazy
Ready1.1s
git clone repo.git
git worktree add ../feature-a
git worktree add ../feature-b
$EDITOR ../feature-a/app.ts

Working tree ready

Hydrates on access, converges in the background.

XFS · reflink-capable fs

Agent working
Copied up frontPlaceholders
Disk used~284 MB

Editing breaks out one file. The rest stay shared, so disk and cache stay flat.

Public aagit installation is not published yet.View release status

Illustrative of the lazy-vs-copy behavior on a measured 16,197-file, 270 MB repo (#10862): plain git copies the whole tree on clone and every worktree add, adding a full 270 MB per worktree; Arivaran Git lays placeholders and adds about 7 MB per worktree on XFS (the figure animated here) and about 19 MB on ext4, in line with the measured 15-39x disk range, and shares one copy in page cache. The lane pacing is illustrative: in the same bench a deferred clone returned in 1.5 s versus 2.7 s for a full clone on XFS, while worktree-add wall time was comparable either way; the measured win is disk and page cache. Reads are byte-correct from the object store while files hydrate. Linux is GA on XFS and ext4 today; macOS and Windows are in beta.

Arivaran dogfood

We learned the copy tax the hard way.

A Claude session put several coding agents in one checkout. Branch operations raced against uncommitted work, so the next run moved each agent into its own worktree.

One honest note carried into memory for next time: I merged the P0 batch with agents but forgot isolation:"worktree", so they raced in the shared worktree (no damage — main is correct and my state intact, but it's the trap my own memory already flags; I'll pass isolation on parallel coding agents going forward).
Internal Arivaran engineering-session note, July 20, 2026

The safe choice had a cost.

Plain Git worktrees isolate each agent, but each checkout materializes another copy of unchanged files.

Keep the isolation. Lose the copy tax.

Arivaran Git keeps separate worktrees and shares unchanged files. A write gets a private inode before it can change another branch.

A short path to an agent-ready worktree.

The live sequence appears only after the public activation contract provides a complete, verified path.

One mechanism, then the measured proof.

Release-gated

Public activation is unavailable. No install command is published until the signed distribution, deterministic verification, and safe rollback contract passes.

  1. Step 1: Install the Linux development build

    On Linux x86_64 with dnf, install the signed development package. It does not enable break-on-write safety.

    curl -fsSL https://get.arivaran.ai/install.sh | sudo sh -s -- aagit
  2. Step 2: MOWI mechanism

    MOWI keeps unchanged paths shared and materializes a private inode before a write.

  3. Step 3: Measured proof

    Measured on a 16,197-file, roughly 270 MB repository across XFS and ext4.

15-39x

less incremental disk per agent worktree

One measured Git result.

The source-build benchmark remains inspectable while public activation is release-gated.

Measured on a 16,197-file, roughly 270 MB repository across XFS and ext4.

Read methodology #10862