VECTA is a workspace for AI coding agents currently in closed testing. The platform allows users to manage tasks via desktop or web browsers, including mobile devices, while the agent executes code on a connected machine. The system architecture includes a runner built with TypeScript and an API based on FastAPI, with the desktop client developed using Electron.
The platform manages project state by creating a snapshot of tracked and untracked files for each task. To ensure accurate diffs, VECTA establishes a baseline using Git worktrees and internal references. This approach prevents existing local changes from being incorrectly attributed to the AI agent's work during the task execution process.
For multi-device workflows, VECTA synchronizes task history and events. When a client reconnects, it uses sequence cursors to fetch missing events from the server, ensuring continuity across sessions. The system enforces access control at multiple levels, including user roles such as owner, operator, and observer, and restricts the agent's environment variables.
The platform addresses security by excluding sensitive paths and files during the preparation of the working copy. It does not support custom Git filters during this process to prevent unauthorized command execution. While VECTA provides tools for task management, the developer notes that it does not implement a custom version control system, relying instead on standard Git mechanisms.
