What's new in Codevalanche.

The page now knows your machine. The download section detects Windows, macOS, and Linux, recommends the matching build, and shows a dedicated note on phones. Buttons stay honestly in preparation until the first installers ship, then activate in place.

Accounts got their own story. A new section explains first-class accounts: subscription runtimes, local models, and more than one plan per provider, with a live usage-handoff illustration.

This changelog went live. Along with a footer language picker showing the 17 languages the app ships in.

Browse the open-model world live. The local catalog now discovers models directly from their upstream source instead of a frozen list, so new releases appear without waiting for an app update.

Fit before you download. Every model is checked against your actual hardware first. A model that will not run on your machine says so before it costs you the gigabytes.

Pull and run on device. Download, start, and talk to a model entirely on your hardware, with the native engine doing the heavy lifting.

Connect more than one plan per provider. A personal and a work Claude subscription can live side by side, each with its own session, identity, and usage view.

Pick the account per task. The account picker pairs any signed-in account with any compatible runtime, so the same objective can run on whichever plan has room.

Groups, projects, and tasks. Work is arranged the way you think about it, and drag and drop puts a conversation where it belongs.

Usage where you type. The composer warns about an approaching plan limit before it interrupts the work, not after.

Import existing agent conversations. Sessions you already ran elsewhere on your machine can be imported and kept in sync next to the projects they belong to.

No silent merges. When an imported conversation continues in two places, Codevalanche forks it instead of guessing which side wins.

Start runs from your phone. The paired companion is no longer read-only: a task can be started away from the desk and reviewed when you return.

Approve diffs from anywhere. Read the change, then approve or decline, while execution and code stay on your machine.