AgentNotch

Source-Aware Design offers immediate visual differentiation between various AI assistants through color-coded indicators. Claude Code activities appear with an orange indicator, OpenAI Codex operations show in blue, and unknown sources display in light blue. This visual distinction allows developers to instantly recognize which AI system is active, particularly useful when working with multiple assistants simultaneously or when troubleshooting integration issues. The color coding extends throughout the interface, ensuring that tool calls, token counts, and other metrics are immediately associated with their originating AI system without requiring manual inspection of source identifiers.

Completion Detection addresses the uncertainty surrounding AI task execution by providing clear notifications when assistants finish their work. This feature monitors activity patterns to determine when an AI agent has completed its current operation, alerting developers through the interface. This eliminates the common problem of wondering whether an AI is still processing or has encountered an error, allowing developers to efficiently manage their workflow between human and AI tasks. The notification system ensures that developers can step away from monitoring while an AI works, confident they'll be alerted when their attention is needed again.

AgentNotch operates through a sophisticated local monitoring architecture that receives telemetry data via the OpenTelemetry Protocol. The application listens on port 4318 for OTLP/HTTP connections, specifically decoding logs from the `/v1/logs` endpoint and metrics from `/v1/metrics`. Developers configure their AI tools to send telemetry to this local endpoint through environment variables or configuration files. For Claude Code, this involves setting the OTEL_EXPORTER_OTLP_ENDPOINT to localhost:4318. For Codex CLI, configuration occurs in the `~/.codex/config.toml` file with specific OTLP exporter settings. Once connected, AgentNotch processes incoming data, extracts relevant tool calls and metrics, and presents them through its native macOS interface.

Concrete use cases demonstrate AgentNotch's practical value across development scenarios. When debugging complex AI-generated code, developers can watch the exact sequence of file reads and edits that led to the final implementation. During code review sessions, teams can verify which files an AI assistant examined before suggesting changes. For cost-sensitive projects, managers can monitor real-time token consumption across multiple developers to stay within budget constraints. Educational settings benefit from visualizing AI thought processes for teaching purposes. The outcomes include reduced debugging time, better cost control, improved team awareness of AI usage patterns, and enhanced ability to audit AI-assisted work for quality assurance purposes.

AgentNotch targets macOS developers using AI coding assistants like Claude Code and OpenAI Codex, particularly those working on MacBook Pro models with notches (2021 and later). The application requires macOS 14.0 (Sonoma) or later and functions as a native Swift application for optimal performance and battery efficiency. While designed for notch-equipped Macs, it gracefully falls back to standard menu bar placement on other macOS systems. The tool is especially valuable for professional developers, engineering teams, technical leads monitoring team AI usage, and organizations concerned with AI development costs. As a 100% local application with no data transmission to external servers, AgentNotch prioritizes privacy while delivering comprehensive monitoring capabilities that transform how developers interact with and understand their AI coding collaborators.