AgentPeek

Product Introduction

1. Definition: AgentPeek is a native macOS menu bar and Dynamic Island (notch) utility application designed for technical monitoring and management of AI coding agents. It falls into the technical categories of developer productivity tools, AI agent observability platforms, and local system monitors.
2. Core Value Proposition: AgentPeek exists to solve the context-switching problem for developers running multiple AI coding agents like Claude Code and Codex concurrently. Its core value is providing real-time, at-a-glance observability and control directly from the macOS menu bar and notch, eliminating the need to constantly alt-tab between terminals, apps, and browser windows to track agent status, token usage, and permission requests.

Main Features

1. Unified Session Overview in Dynamic Island/Menu Bar: This feature provides a live, consolidated view of all active Claude Code and Codex sessions. It works by installing lightweight local hooks that read session files and system activity. The technology parses agent data to display session status (working, waiting, finished), project context, and aggregates them into a compact, always-visible interface within the Mac's notch or menu bar, enabling instant visual triage.
2. Real-Time Token Usage and Rate Limit Tracking: AgentPeek monitors and visually represents the 5-hour and 7-day API usage limits for Claude and Codex agents in real time. It works by intercepting and parsing the usage data reported by the agents themselves. The feature includes live countdown timers to reset, providing proactive alerts to prevent mid-task rate limit interruptions, a critical function for maintaining developer workflow continuity.
3. In-Line Permission and Prompt Management: When an AI agent requires user input—such as approving a plan, granting a file system permission, or answering a question—the request surfaces directly within the AgentPeek interface. This works via the same local hooks that detect pending agent states. It allows for keyboard-driven responses (e.g., ⌘A to Allow, ⌘N to Deny) without requiring mouse interaction or window focus change, drastically reducing workflow disruption.
4. Integrated Local Development Server Dashboard: Beyond AI agents, AgentPeek extends monitoring to local development servers. It detects running servers, displaying their runtime (e.g., Node.js, Python), port number, and associated project. The technology likely uses process and network socket enumeration. This allows developers to stop any local server with a single click from the unified interface, consolidating dev environment management.
5. Quick-Access Contextual Controls and History: This feature provides one-click navigation from the notch/menu bar to key agent components: skills, plugins, Model Context Protocol (MCP) servers, configuration files, and logs. It also maintains a plain-language, timestamped history of all tool calls made by the agents, offering a summarized audit trail of what files and operations were touched during a session.

Problems Solved

1. Pain Point: Severe workflow fragmentation and loss of focus caused by managing multiple, silent AI agent terminals that require constant manual checking for status updates, token limits, and permission prompts.
2. Target Audience: The primary user personas are software engineers, full-stack developers, and technical leads who extensively use AI coding assistants like Claude Code and Codex for iterative coding, refactoring, and debugging across multiple projects simultaneously. Secondary users include DevOps engineers and engineering managers seeking better observability into AI-assisted development workflows.
3. Use Cases: Essential scenarios include: a developer running parallel agent sessions to debug a backend API and a frontend component simultaneously; an engineer approaching their hourly Claude token limit while in a deep "flow state" who needs a proactive warning; a developer who has background agents running long tasks (e.g., test generation, documentation) while focusing on primary work in their IDE or browser.

Unique Advantages

1. Differentiation: Unlike generic terminal multiplexers (tmux, screen) or menu bar system monitors, AgentPeek has semantic understanding of AI agent workflows. It doesn't just show terminal output; it interprets agent state, token economics, and permission flows specific to Claude and Codex. Compared to checking individual agent UIs or CLI outputs, it aggregates all critical data into a single, persistent visual layer.
2. Key Innovation: The product's core innovation is its non-invasive, hook-based local monitoring architecture that requires zero changes to existing developer workflows. It leverages the standardized file and configuration locations used by Claude Code and Codex to provide deep observability without being an API proxy or requiring a subscription to a cloud service. The integration of this data into macOS's native Dynamic Island is a unique platform-specific UX innovation for developer tools.

Frequently Asked Questions (FAQ)

1. Is AgentPeek safe, and does it send my code or prompts to the cloud? AgentPeek operates entirely locally on your Mac. Your session transcripts, code diffs, prompts, and token usage data never leave your machine. The application has no analytics or telemetry. The only external network calls are for license validation and update checks via the Sparkle framework, which transmit only a machine ID, app version, and license key if purchased.
2. How does AgentPeek work with Claude Code and Codex without requiring configuration changes? AgentPeek works by installing small, local hooks that read from the standard file paths and configuration stores used by Claude Code and Codex on macOS. Because these agents write session data, logs, and state to predictable local locations, AgentPeek can monitor them regardless of whether they are run in a terminal (iTerm, Warp, default Terminal) or via their official desktop applications, requiring no user reconfiguration.
3. What are the system requirements for running AgentPeek? AgentPeek requires an Apple silicon Mac (M1, M2, M3, or later chips) running macOS Sonoma (14.0) or a newer version. It is a native Swift application optimized for ARM architecture and utilizes modern macOS APIs like the Dynamic Island, which is why Intel Macs and older OS versions are not supported.
4. What does the one-time $15 license include? The one-time $15 license fee grants a perpetual license for AgentPeek on your Apple silicon Mac. This includes access to all current features, support for unlimited concurrent AI agent sessions, and all future feature updates and bug fixes. There is no subscription, recurring fee, or account required.
5. Can I try AgentPeek before purchasing a license? Yes, AgentPeek offers a full-featured, free trial for 2 days upon first launch. This allows you to integrate it into your actual development workflow to evaluate its utility. After the trial period, a one-time purchase is required to continue using the software.