Summary

  • Pervaziv AI announced Cortex Cloud on September 10, adding a managed remote destination alongside local execution.
  • The company says project-aware handoffs use a stable committed revision and will not silently move cloud-selected work back to a local machine.

A developer can close a laptop without deciding which unfinished edits an AI task should inherit. Cortex Cloud, announced by Pervaziv AI on September 10, makes that distinction central to its managed-execution pitch. The remote task is meant to outlast one open client, but its starting project state must be deliberately established.

This is a change in execution location, not the first appearance of cross-device continuity in Cortex. The company's August 25 Cortex Connect announcement already linked requests, progress and results across mobile, browser and Visual Studio Code. Project-aware work then remained centred on the connected development workspace, where code and tools were available. September's addition gives eligible work another destination.

According to the launch description, users choose local or managed cloud execution from an authenticated Cortex session. For project-aware cloud work, the handoff uses a stable committed source revision. Uncommitted local edits are not silently assumed to exist remotely. If the working copy is unsuitable, the handoff stops so the user can prepare it.

That refusal has an economic consequence as well as a control function. Remote capacity cannot remove the need to decide what the task is actually allowed to act on. Somebody still has to choose the revision, resolve the preparation issue and connect the returned work to that starting point. A more convenient request surface does not make this work disappear.

The destination choice also survives a failed admission. Pervaziv AI says that when the cloud path cannot proceed, Cortex reports the relevant state rather than silently falling back to local execution. Once eligible, work runs in a managed workspace associated with the authorised user, project and workflow, subject to organisational controls and availability.

The company's durable-state design preserves progress and outcomes so an active screen is not their sole owner. Its operational view can show bounded metadata such as state, timestamps, region and resource use, rather than routinely exposing raw code or conversations. These are supplier descriptions, not an independent recovery or privacy audit. They do not establish that customer code never reaches the execution environment, or that interrupted external actions can never occur twice.

The earlier Connect release matters here. It positioned mobile and browser access as a way to carry intent into an existing workspace, not as a requirement to duplicate that environment elsewhere. Cloud execution adds an option to that model; it does not prove that every task, local dependency or permission can move unchanged.

Commercial terms need a similarly careful separation. Pervaziv AI's public pricing page describes general per-user packages, project limits and feature tiers. The examined material does not turn those packages into a complete Cortex Cloud compute quotation. It does not specify the cloud unit rate, included runtime or retry accounting needed to calculate a particular team's remote-execution bill.

For buyers, the relevant unit is therefore an admitted, reviewable piece of work, not simply a seat that can send prompts from several devices. The possible benefit is less dependence on a workstation staying open. The corresponding dependency is a managed service that must admit the right revision, explain blocked work and return an intelligible outcome.

Sources