Superagent

Product Introduction

1. Definition: Superagent is a locally-hosted, open-source macOS desktop application that functions as an integrated development environment (IDE) for AI coding agents. It provides a persistent, visual workspace where agents like Claude Code can operate with direct access to system resources.
2. Core Value Proposition: It exists to transform AI coding assistants from ephemeral, terminal-based chatbots into persistent, context-aware collaborators with a visual interface. Its core value is giving AI agents a "computer" — including a real browser, iOS Simulator, and file system access — while keeping all work local, private, and integrated into the developer's existing workflow without requiring cloud accounts or servers.

Main Features

1. Persistent, Task-Based Sidebar Chats: Each development task exists as a dedicated, persistent chat session within a sidebar. These sessions maintain full context, survive application restarts, and run independently. The interface visually indicates agent status (working, finished, unread), allowing developers to manage multiple concurrent agent-driven tasks without context loss.
2. Isolated Git Worktree Integration: Every chat session can be launched on its own isolated Git worktree. This allows the AI agent to make commits and changes on a separate branch without touching the developer's primary working directory. It ensures a clean git status on the main branch while the agent works autonomously on fixes or features.
3. Native Browser Automation & Session Persistence: The agent operates a real, instrumented browser instance (likely based on Chromium) that retains the user's logged-in sessions (e.g., Gmail, admin dashboards). The agent can perform complex, multi-step workflows—reading, clicking, typing, navigating—within the user's authenticated context, automating tasks beyond code editing.
4. Integrated iOS Simulator with Framebuffer Streaming: Superagent includes direct control over a macOS iOS Simulator, streaming its framebuffer into the application window. The AI agent can boot the simulator, launch apps, and perform precise UI interactions (tap, type, swipe) for testing and debugging mobile applications, eliminating the need for manual simulator management.
5. Dual-Viewport Live Preview: Developers can view any web page simultaneously in both a desktop and a mobile phone viewport within the same browsing session. Both views are live and interactive; scrolling or changes in one are reflected in the other, enabling efficient cross-viewport responsive design testing.
6. End-to-End Encrypted iPhone Companion: A paired iPhone app allows remote monitoring and control of the Mac-based agent. Notifications, live chat transcripts, and agent approval requests are sent via a direct, end-to-end encrypted connection through an optional, open-source relay. This enables remote workflow management without data exposure.
7. Multi-Agent Desktop Workspace: The application provides a computer desktop-like environment where multiple AI agent instances can work simultaneously in movable, resizable windows. This "Computer" view offers a holistic dashboard of concurrent agent activities, moving beyond a single chat interface to a multi-agent orchestration surface.

Problems Solved

1. Pain Point: The disjointed, stateless nature of typical AI coding assistants that lose context between sessions, lack integration with local development environments, and cannot perform visual or browser-based tasks.
2. Target Audience: Software engineers and developers (particularly full-stack and mobile), QA engineers, product managers, and technical founders who use AI assistants for coding, testing, debugging, and repetitive web-based tasks. It targets users who prioritize privacy, local execution, and deep workflow integration.
3. Use Cases:
   • Context-Preserving Code Reviews & Fixes: Investigating a flaky test in an isolated branch while continuing primary development uninterrupted.
   • Automated Cross-Platform UI Testing: Instructing an agent to verify a button flow works on both a desktop browser and an iOS simulator.
   • Authenticated Web Workflow Automation: Having an agent log into a web dashboard, extract data, and generate a report, or handle routine customer outreach via a logged-in email client.
   • Responsive Design Implementation: Writing and previewing CSS changes with immediate feedback on both desktop and mobile viewports live.
   • Asynchronous Development Pairing: Starting a complex refactor task with an agent, closing the laptop, and later reviewing commits and progress from an iPhone.

Unique Advantages

1. Differentiation: Unlike cloud-based AI platforms (e.g., Cursor, GitHub Copilot Chat) or simple CLI wrappers, Superagent runs entirely locally, owns its full UI/UX, and provides agents with direct access to native OS capabilities (browser, simulator, filesystem). It competes with agent frameworks but bundles the environment, persistence, and interface into a single, user-friendly desktop app.
2. Key Innovation: The "Persistent Visual Workspace" model. It re-conceptualizes the AI agent from a conversational interface into a resident worker with dedicated resources (browser, simulator, git worktree). The integration of local, authenticated browser automation and isolated git worktrees is a technically sophisticated combination that bridges the gap between AI instructions and tangible, context-aware execution on a developer's machine.

Frequently Asked Questions (FAQ)

1. Is Superagent a cloud service or does it run locally? Superagent is a 100% local macOS application. There is no Superagent cloud server; all code execution, browser sessions, chat data, and file changes occur on your local machine. The only optional external component is a privacy-focused relay for iPhone pairing, which you can also self-host.
2. How does Superagent handle my browser passwords and logged-in sessions? It uses a dedicated browser profile stored locally on your Mac, identical to how standard browsers like Chrome or Edge manage sessions. Your logins and cookies never leave your computer. The AI agent interacts with this local profile, so it only has access to sessions where you are already authenticated.
3. Can I use Superagent with AI models other than Claude? The application is designed to be model-agnostic. While it currently integrates with Claude Code, the architecture supports other models. The roadmap explicitly mentions future compatibility with Codex and Antigravity, and the open-source nature allows for community integration of additional AI backends.
4. How does the Git worktree feature work, and why is it useful? When you start a chat for a specific task, Superagent can create a Git worktree—a separate directory linked to a new branch. All agent commits happen there. This keeps your main working directory completely clean and allows you to context-switch instantly, merging the agent's work via a standard pull request when ready.
5. Is Superagent truly free and open source? Yes. The desktop app (macOS), iPhone app, and relay server are all released under the permissive MIT license on GitHub. You can download pre-built, Apple-notarized binaries for free, audit the code, modify it, or build it yourself without any licensing cost. It uses your existing AI model subscription (e.g., with Anthropic).