AgentDock

Product Introduction

1. Definition: AgentDock is a macOS-native desktop application (DeveloperApplication) designed as a unified management and interaction hub for AI-powered coding agents. It functions as a system-level overlay that monitors, aggregates, and facilitates interaction with multiple concurrent AI coding sessions.
2. Core Value Proposition: It exists to solve the context-switching and management overhead for developers running multiple AI coding agents (like Claude Code, Codex, and others) simultaneously. Its primary value is providing a single, glanceable interface to monitor agent status, centrally manage agent queries, and interact with any agent from anywhere within the macOS workflow, thereby increasing developer productivity and reducing cognitive load.

Main Features

1. Unified Agent Status Dock: The app displays all active AI coding agents as color-coded dots on the edge of the user's screen (blue for working, amber for needing input, green for finished). This visual layer is drawn system-wide, providing instant, at-a-glance awareness of all concurrent AI tasks without requiring window switching. It works by reading the local session files generated by the supported agents and terminal emulators.
2. Centralized Inbox for Agent Queries: All questions, permission prompts, and decision points from every monitored agent are aggregated into a single, keyboard-navigable Inbox (⌃⌥I). Users can review and respond with a single keystroke (1-9). The app then types the selected response directly into the correct terminal session (via integration with cmux, tmux, iTerm, WezTerm, kitty, or herdr) after a 2-second undo window. For unsupported terminals/editors, it leverages Claude Code's native hook system to forward answers.
3. System-Wide Agent Messaging (⌥⌥): A global hotkey (press Option twice) allows users to send a message to any active agent from within any application. The command includes auto-complete; starting a message with an agent's name and a colon (e.g., shop-web:) pre-selects the target. This enables quick, contextual instructions without finding and focusing the specific terminal window.
4. Local AI-Powered Work Recaps: Upon an agent completing a task, AgentDock generates a concise, one-line recap and suggests three potential next steps. Critically, this summary is generated locally on the Mac using Apple Intelligence (on-device AI), ensuring code and context never leave the machine. Users can instantly send a suggested next step with ⌘1, ⌘2, or ⌘3.
5. Detailed Agent Hover Cards & Analytics: Hovering over a status dot reveals a card with technical details: the AI model in use (e.g., Opus 5.5), current turn timer, project context (directory), and token usage statistics for the current day. Token counts are derived from the agents' own local session logs and distinguish between new tokens and cache re-reads, providing insight into computational cost.

Problems Solved

1. Pain Point: Fragmented AI Agent Management. Developers using multiple AI coding agents suffer from tab/window overload, constant context switching to check status or answer prompts, and difficulty tracking the output and progress of parallel tasks.
2. Target Audience: Professional and Prolific Developers & Engineers. Specifically, macOS-based software engineers, full-stack developers, and technical leads who routinely employ multiple AI coding assistants (Claude Code, Codex, Gemini via herdr) across several projects or terminals simultaneously to accelerate coding, debugging, and testing workflows.
3. Use Cases:
   • Parallel Feature Development: Managing Claude Code working on a frontend React component while Codex builds a backend API endpoint, with both agents asking for design decisions.
   • Cross-Project Maintenance: Monitoring bug fixes in a legacy iOS project (notes-ios) while simultaneously having an agent research and draft documentation for a new web service (shop-api).
   • Asynchronous Work Review: Stepping away from the desk and using the iPhone web interface to monitor agent status and answer critical permission prompts that would otherwise block progress.

Unique Advantages

1. Strengths & Limitations (Pros & Cons):

   • Pros:
     • Unparalleled Workflow Integration: Deep, direct integration with major terminal multiplexers (tmux, cmux) and emulators allows for genuine bidirectional communication (reading status and typing input), which is unique.
     • Strict Privacy & On-Device Processing: No account, no cloud analytics, no data exfiltration. Status is read from local files; recaps use on-device Apple Intelligence. This is a major security and privacy advantage.
     • One-Time Purchase Model: A $19 perpetual license for version 1.x is a significant cost advantage compared to subscription-based developer tools.
     • Lightweight & Native: A 5.1 MB, notarized Mac app that operates as a efficient menu bar utility.
   • Cons:
     • Platform & OS Lock-in: Exclusively for macOS 26+, limiting its user base to developers on the latest Apple operating systems.
     • Dependent on Agent/App Support: Full functionality requires using supported terminals or agents. While Claude Code's hook provides broad baseline support, the powerful "direct typing" feature is limited to specific terminal apps.
     • Niche Audience: Its value is directly proportional to the number of AI coding agents a developer runs concurrently. For users of a single agent, it may be overkill.

2. Key Alternatives & Differentiation:

   • Manual Terminal/Editor Tab Management: The default alternative is simply juggling multiple terminal tabs or editor windows.
     • Differentiation: AgentDock eliminates constant tab hunting and provides a unified status layer and input queue, reducing cognitive load and saving time.
   • Terminal Multiplexer Panes (tmux/cmux): While these allow viewing multiple sessions, they lack aggregated status overviews, a unified query inbox, and system-wide interaction capabilities.
     • Differentiation: AgentDock adds an intelligent management layer on top of multiplexers, specifically designed for the interaction patterns of AI agents, not just shell sessions.
   • Individual Agent Desktop Apps (Claude.app, Codex.app): These run in isolated windows.
     • Differentiation: AgentDock is agent-agnostic, providing a single control plane across different AI providers and interfaces (terminal vs. desktop app), which the native apps cannot do.

Frequently Asked Questions (FAQ)

1. Does AgentDock work with GitHub Copilot or Cursor's agent? AgentDock primarily integrates with agents that run in terminal sessions or have accessible local session files, like Claude Code and Codex. While it can detect agents within editors like VS Code, Cursor, or Zed through Claude Code's hook, its ability to interact with deeply integrated editor plugins like GitHub Copilot Chat is limited to status visibility; it cannot type into those proprietary editor interfaces.
2. Is my code or conversation data sent to the developer's servers? No. AgentDock operates on a strict local-first, privacy-by-design principle. It reads only the session log files already generated on your Mac by the AI agents themselves. The AI-generated recaps use the on-device Apple Intelligence framework. No code, prompts, or conversation data is uploaded to any external server, including the developer's.
3. What happens after the 7-day free trial ends? The core monitoring features—the status dots on the screen edge and the informational hover cards—continue to work indefinitely for free. The premium features that require a $19 license are: actively answering questions from the centralized Inbox, using the ⌥⌥ global messaging shortcut, and accessing the agent interface from your iPhone. This "freemium" model allows continuous basic monitoring.
4. How does AgentDock impact system performance or battery life? As a lightweight, native Mac app that primarily polls local log files and acts as a system overlay, AgentDock has a minimal performance footprint. It does not run continuous heavy processing. The most resource-intensive operation—generating recaps—leverages the dedicated Apple Intelligence processor (Neural Engine) on Apple silicon Macs, designed for efficient on-device AI tasks with minimal battery impact.
5. Can I use AgentDock with agents running on remote servers or in Docker containers? AgentDock's core functionality depends on accessing local session files and, for full features, directly interacting with local terminal sessions. It is not designed to monitor agents running inside remote SSH sessions, Docker containers, or cloud development environments where it cannot access the necessary local process interfaces and log files. Its use case is focused on locally-hosted AI coding agents.