Zenborg User Guide

Zenborg is a desktop application that runs and coordinates multiple AI coding agents on your local machine.

Most AI coding assistants work alone. They write code, test their own changes, and declare tasks done. When an agent hallucinates or makes a mistake, you must find and fix the bug yourself.

Zenborg changes this workflow. It puts agents into coordinated teams where they plan, implement, and review each other’s work before any changes touch your main branch.

What Zenborg Is

Zenborg is a local orchestration engine for macOS and Windows. It connects to the coding agents already installed on your computer and coordinates them through structured workflows called Agent Loops.

In Zenborg:

One agent drafts an implementation plan.
Two independent agents review the plan and spot architectural risks.
A synthesis step merges feedback into an approved plan.
A worker agent writes code in an isolated Git worktree.
Reviewer agents inspect the diff and request targeted fixes.
Automated fix rounds correct issues before you inspect the result.

Your project files are never edited directly during a run. All changes happen in disposable Git worktrees created specifically for each loop.

Who It Is For

Zenborg is built for software engineers and technical teams who already use AI coding tools, including:

Claude Code (claude) from Anthropic
Google Antigravity (antigravity / agy) and Gemini CLI (gemini)
Hermes (hermes)
OpenCode (opencode)
Agent Client Protocol (ACP) tools like Cursor CLI, GitHub Copilot CLI, and Kiro CLI
Gemini Web and ChatGPT Web subscriptions in the browser

You do not need to be an expert in multi-agent systems. If you can run a CLI command or prompt an AI model, you can run Zenborg.

How Zenborg Saves Time and Money

1. Runs on Subscriptions You Already Pay For

Zenborg does not charge per-token fees or mark up API rates. It connects directly to your existing CLI logins and browser subscriptions. If you pay for Claude Pro, ChatGPT Plus, or Gemini Advanced, Zenborg uses those active accounts.

2. Browser Workers for Zero-Token Tasks

Zenborg includes dedicated browser workers for Gemini Web and ChatGPT Web. Heavy tasks like draft planning, synthesis, and code reviews can run in the browser without consuming API tokens or billable API credits.

3. Automated Error Detection

Single agents often repeat their own mistakes. When different models critique each other, they catch bugs, missing edge cases, and invalid assumptions early. You spend less time debugging failed agent runs.

4. Background Concurrency

Zenborg manages machine load automatically. You can start a loop, switch tasks, and let Zenborg run planning, reviews, and test runs in the background.

Documentation Map

Explore the user guide using the topics below:

Getting Started

macOS Installation: Download, install, launch, and activate Zenborg on macOS.
Windows Installation: Install on Windows, configure permissions, and manage processes.
Connect Your Agents: Detect and authenticate CLI and ACP coding agents.

Core Concepts

Agent Loops: Learn the 7-stage pipeline, worktree isolation, fallback chains, and concurrency controls.
Session Chats: Run multi-agent team chats, use @mentions, and manage conversation handoffs.
Projects and Goals: Organize workspaces, configure target paths, and track high-level goals.
Browser Workers: Use Gemini and ChatGPT web interfaces in a dedicated Chrome profile.
Skills and MCP: Understand the 15-rule skill security scanner and loop MCP isolation.

Operations and Support

Diagnostics and Troubleshooting: Resolve port conflicts, stale locks, provider errors, and memory limits.
Plans and Pricing: Compare Free, Beta, and Paid tiers, including entitlement limits.
Frequently Asked Questions: Quick answers to common questions about safety, Git, and keys.

Step-by-Step Tutorials

Your First Agent Loop in 10 Minutes: Create a goal, configure manager and worker agents, and review the output.
Run a Loop on Your Gemini Subscription: Log in to Gemini Web and execute tasks with zero API fees.
Multi-Agent Review Team: Configure a multi-agent team with cross-agent code review and automatic fallback.