AgentPeek

Product Introduction

1. Definition: AgentPeek is a local-first macOS menu bar application (a "notch" utility) designed for real-time monitoring and management of Anthropic's AI coding agents, specifically Claude Code and Claude Codex. It falls into the technical categories of developer productivity tools, AI agent observability platforms, and system tray utilities for macOS.
2. Core Value Proposition: AgentPeek exists to solve context-switching overhead for developers using AI coding agents by providing instantaneous, glanceable access to agent sessions, permissions, local development servers, and token usage metrics directly from the macOS menu bar, eliminating the need to constantly alt-tab between terminals and IDEs.

Main Features

1. Mac Notch Session Dashboard: Provides a persistent, system-level overlay displaying all active Claude Code and Claude Codex sessions. The interface shows session names (e.g., "refactor auth middleware"), the associated terminal or editor (e.g., Ghostty, VSCode), and the local project directory. Clicking any session instantly focuses the application window where that agent is running.
2. Quick Routes for Agent Folders: Allows developers to pin frequently accessed project directories. With one keystroke triggered from the menu bar, users can instantly open a terminal or file explorer at the root of any pinned project, enabling rapid navigation to the exact context of an AI agent's work. This feature directly integrates with the session dashboard for seamless workflow.
3. Local Dev Server Management: Actively detects locally running development servers on ports 3000 through 9999. For each detected server (e.g., localhost:5173 for a Vite/Node app, localhost:3000 for Next.js), it displays the runtime, technology, and provides actions to open the server in a browser, copy the URL, or stop the server—all from the menu bar without switching contexts.
4. Dual-Pane Token Usage Analytics: Delivers real-time and historical visibility into AI agent resource consumption. It displays side-by-side, glanceable metrics for both Claude Code and Claude Codex across two time windows: a rolling 5-hour window for immediate usage and a 7-day window for longer-term trend analysis (e.g., "5H 17%", "7D 70%").
5. Local-First, Privacy-Centric Architecture: All monitoring data, including session history and token usage, is processed and stored exclusively on the user's local machine (Apple silicon Mac). No code, conversation data, or analytics are transmitted to external servers, ensuring complete data privacy and security for proprietary development work.

Problems Solved

1. Pain Point: Severe developer workflow fragmentation and context-switching fatigue caused by constantly toggling between an IDE/terminal running an AI agent and other applications, leading to lost focus and reduced productivity.
2. Pain Point: Lack of immediate visibility into AI agent state, requiring manual checks to see if an agent is waiting for user permission, to track token expenditure, or to manage multiple concurrent agent sessions.
3. Pain Point: Inefficient navigation between multiple active development projects and their corresponding local servers, requiring repetitive terminal commands or IDE project switching.
4. Target Audience: Professional software engineers, full-stack developers, and technical leads who regularly use Claude Code or Claude Codex within terminals or AI-native IDEs (like Cursor, Windsurf) on macOS. The primary persona is a developer working on multiple repositories who needs to maintain oversight of AI-assisted coding sessions.
5. Use Cases: Monitoring a long-running Claude Code refactoring session while writing documentation in another window. Quickly granting approval when Codex requests permission to execute a command. Comparing the token consumption of Code versus Codex on a specific feature branch. Managing a local React dev server, a Go API server, and a documentation preview server simultaneously from a central hub.

Unique Advantages

1. Differentiation: Unlike built-in IDE plugins or terminal logs, AgentPeek offers a unified, system-level observability layer that works across any terminal emulator (iTerm, Warp, Ghostty) and editor, providing a consistent interface regardless of the user's toolchain. It is not tied to a specific editor like Cursor.
2. Key Innovation: Its deep integration with the macOS menu bar as a primary interface is its core innovation. It transforms a typically passive system area into an active command center for AI-assisted development, enabling sub-second interactions (jump to session, manage server) that are faster than traditional application switching.
3. Business Model: A one-time purchase license ($9) for lifetime access and future updates contrasts sharply with the prevalent SaaS subscription model, offering exceptional long-term value and cost certainty for individual developers.

Frequently Asked Questions (FAQ)

1. Does AgentPeek work with Cursor, Windsurf, or other AI coding tools? Yes, AgentPeek is compatible with any AI coding tool that runs Claude Code or Claude Codex in the background, including Cursor, Windsurf, and traditional terminals. It monitors the agent processes directly, not the specific IDE.
2. Does AgentPeek access my code or AI conversation data? No. AgentPeek operates on a strict local-first principle. It reads system-level process information and metadata (like session names, ports, and token counts) but does not access, read, store, or transmit your proprietary source code or the content of your conversations with Claude.
3. What are the system requirements for AgentPeek? AgentPeek requires a Mac with Apple silicon (M-series chip) running macOS Sonoma (14.0) or later. It is optimized for the native performance and architecture of modern Apple hardware.
4. How does the license and free trial work? AgentPeek offers a full-featured, 2-day free trial. After the trial, users can make a single one-time payment of $9 for a lifetime license, which includes the current version and all future updates, with no subscriptions or recurring fees.
5. How does AgentPeek detect when an agent needs permission? The application monitors the state of the Claude Code and Codex processes. When an agent pauses execution to request user approval (e.g., to run a command or modify a file), this state change is detected and prominently indicated in the menu bar session list, alerting the developer to take action.