AI Capabilities

What ZeroSphere's agent can actually do, organized by area.

Reading and Understanding Code

The agent reads your full codebase before acting. It sees the entire project, not just the file you have open.

File operations:

  • Read any file in your workspace
  • List directories and understand project structure
  • Search files by name or content
  • Find functions, classes, and patterns across the codebase

Code understanding:

  • Traces dependencies between files
  • Understands how changes in one file affect others
  • Reads compilation errors and test output as context

Writing and Editing Code

What it can do:

  • Create new files and directories
  • Edit existing code with precision
  • Refactor across multiple files in a single task
  • Maintain your existing code style and formatting

What it will not do:

  • Touch files outside your workspace
  • Modify files matching your .agentignore patterns
  • Run destructive operations without permission

Running Commands

The agent executes shell commands and reads their output to decide what to do next.

Typical command usage:

  • Build commands and compilation
  • Test suite execution
  • Dev server startup
  • Background process monitoring

Every command goes through the permission system. You approve what runs.

Launching and Testing Your App

This is what separates ZeroSphere from every other IDE.

The agent creates a virtual display session — an isolated display your OS treats as a real screen. Your app launches on it. The agent watches using raw pixel capture. No DOM. No accessibility tree. No element IDs.

What it does inside the virtual display:

  • Launches desktop apps, dev servers, browsers, Electron apps
  • Clicks, types, scrolls, drags, and uses keyboard shortcuts
  • Reads the screen to understand what happened
  • Catches visual failures and traces them back to code
  • Fixes what broke without you describing anything

Session details:

  • Fully isolated from your main desktop
  • You see the live feed in the agent panel
  • You can take control or stop it at any point
  • Sessions expire after 20 minutes of inactivity and clean up automatically

Window management:

  • Finds windows that opened on the wrong display
  • Moves them into the correct session
  • Can also operate your real primary display directly for apps that cannot run in a virtual display

See Virtual Display and Window Tools for details.

Web Search

When TAVILY_API_KEY is set, the agent can search the web for up-to-date information — library docs, error messages, changelogs, or anything else it needs that's not in your codebase.

Version Control

The agent understands Git and works with your existing workflow:

  • Reads current branch and status
  • Understands what has changed and why it matters for the task
  • Can stage, commit, and manage changes when instructed

Task and Progress Tracking

For complex multi-step tasks the agent tracks its own progress:

  • Breaks large tasks into ordered steps
  • Marks steps complete as it goes
  • Shows you where it is in the task at any point
  • In Planning mode, saves full task lists and a completion report when done

What the Agent Cannot Do

  • Access files outside your workspace (unless you enable it in permissions)
  • Connect to the internet without Tavily configured
  • Act outside your permission settings
  • Retain context between sessions (unless memory saved)