uzairansaruzi
Featured
hermex
Hermex is a native SwiftUI iPhone app that lets you control a self-hosted Hermes AI agent directly from your phone, with no subscriptions, tracking, or third-party relay.
Why it is worth attention
It offers a zero-config, no-API, no-account way to visually monitor AI agent sessions in the terminal, uniquely integrating with the MacBook notch for glanceable status.
Who it is for
- Mac users with notched MacBooks
- Developers using Claude Code or Codex
- Terminal power users
- AI agent workflow enthusiasts
Use cases
- Monitoring multiple agent sessions running simultaneously in the terminal
- Quickly seeing when an agent finishes via a green indicator
- Accessing session history with prompts and subagents through a clickable panel
- Customizing the Codex pet animation by switching spritesheets
Strengths
- No configuration required—no hooks, APIs, or accounts needed
- Real-time polling every 3 seconds with process and file descriptor detection
- Transparent, click-through window that never blocks other apps
- Supports multiple terminal apps and pet customization
Considerations
- ~30 second afterglow in idle/busy detection due to bursty transcript writes
- Requires macOS 12+ and a notched MacBook (or a virtual notch on other displays)
- Only monitors Claude Code and Codex agents, not other AI tools
README quick start
Agent Notch
Your AI agents, living next to the MacBook notch.
While Claude Code or Codex is working, its mascot walks beside the notch — the Claude Code banner critter for Claude, the official Codex pet for Codex. Each agent has its own slot: the moment one finishes, its mascot becomes a green blob (even while the other keeps working). The green is a "finished since you last looked" notification — focusing your terminal clears it. Click to open a panel of your sessions, grouped by prompt, with every subagent tucked under a dropdown.
The panel
One row per session, titled by the tool, led by your latest prompt — not the agents' chatter. Subagents (Codex's philosopher swarm, Claude's Task agents) fold under a ▸ N subagents dropdown. Running rows show their mascot walking in place; finished rows get a green pixel checkmark. The tag on the right is the actual model that session runs.
How it works
No hooks, no APIs, no accounts. Liveness detection follows open-vibe-island's model — a session is a running agent process in a terminal — polled every 3 s:
psfindsclaude/codexprocesses attached to a TTY (headless/background sessions are ignored)lsofmaps each process to the transcript it holds open (Codex), or to its working directory (Claude Code, which doesn't keep the transcript fd open)- transcripts provide the metadata: prompts, snippets, models, subagents
- Claude Code:
~/.claude/projects/*/*.jsonl(+/subagents/agent-*.jsonl) - Codex:
~/.codex/sessions/**/*.jsonl, grouped byparent_thread_id
- Claude Code:
Within a live session, busy vs idle is a hybrid: process alive + transcript written in the last 30 s = busy (mascot walks); alive but quiet = idle (nothing in the notch, dimmed row in the panel); process gone for 2 polls = done (green blob). Sessions idle over 6 h drop off the panel. Activating a terminal app (Ghostty, Terminal, iTerm2, kitty, Warp, Alacritty) acknowledges finished agents and clears their green indicator.
Known limitation: the ~30 s afterglow
Busy/idle is inferred from transcript write times, and transcript writes are bursty — so the mascot keeps walking for up to ~33 s (30 s window + 3 s poll) after a turn actually ends, and conversely a turn's quiet stretches are smoothed over. No process-level proxy (network, CPU, child processes) can
Description
The open-source alternative to vibe-island
Related repositories
Similar projects matched by category, topics, and programming language.
Blaizzy
Featured
nativ
Nativ is a native macOS app that lets you run AI models locally on Apple silicon, offering chat, model management, performance analytics, and an OpenAI/Anthropic-compatible API server.
simonw
Featured
speech-analyzer-cli
A macOS CLI tool that transcribes prerecorded audio using Apple's on-device speech APIs and outputs plain text, JSON, JSONL, SRT, or WebVTT.