Image: GitHub
Why it mattersA developer running several coding agents at once no longer opens each one to find out which is still working. The HUD puts all of them on the edge of the screen.
The developer jazzenchen has released Agent HUD. It is an open source app that shows the quotas, tokens and sessions of 18 coding agents in one place. The repository is Apache 2.0 and has 85 stars at seven days old. Agent HUD is a cross platform port of the author's earlier native Swift app for the Mac. The port is built with Tauri 2. The core is Rust. The interface is React. A small native layer handles each operating system.
What it shows, and where
Agent HUD runs on the Mac today. Windows and Linux are in preview. iPhone and Apple Watch pair with a Mac through iCloud.
The author lists 18 supported agents. Claude Code and Codex lead the list. The rest fall in two groups. First group: DeepSeek Harness, Antigravity, Cursor, Grok CLI, Grok Bot. Second group: GitHub Copilot CLI, OpenCode, Kimi, Pi, OpenClaw, Hermes Agent, ZCode, CodeBuddy, WorkBuddy, Qwen Code and Devin CLI. Agent HUD also reads the GLM Coding Plan quota and the Qoder builds. The app finds each agent on its own once the user signs in to it.
On the Mac the HUD sits at the notch by default. The author also offers a Simple mode for any screen edge. A Dynamic mode puts each agent's logo in a row with a glow behind it. The author writes that one glow colour per agent is green under 70 percent used. It turns yellow from 70 percent. It turns red from 90 percent. Hover opens a card with each quota's reset time or burn rate. One click jumps back to the session's thread, tab or terminal pane.
Token spend and approvals
The author writes that Agent HUD tracks tokens over the last 5 hours, 24 hours, 7 days and 30 days. The bars run from 15 minutes to a day. History stays in a local SQLite database on the computer for 30 days. The HUD breaks token spend down by input, output, reasoning, cache writes and cache reads. Each session shows its turns, calls, context fill and cache hits. A cost estimate at each vendor's list price appears next to the token count, marked with an "approximately" symbol.
Approvals, questions and replies live on the HUD too. The author writes that nine agents route approvals through the HUD. The list is Claude Code, Codex, DeepSeek Harness, Antigravity, ZCode, CodeBuddy, WorkBuddy, Qwen Code and Devin CLI. Codex approvals need the HUD's hook trusted inside Codex. DeepSeek Harness approvals need the HUD's plugin.
iPhone and Apple Watch
The author writes that iPhone and Apple Watch show the Mac's quotas, tokens and sessions without a subscription. The phone and the Mac must share a network. A separate subscription called Agent HUD Remote adds the same view on any network, through the user's own iCloud. Remote also lets the phone read Claude Code, Codex, Grok CLI, Grok Bot and Antigravity sessions. The phone can approve a tool call, answer a question and approve a plan. A note spoken to the Apple Watch becomes a new session on the Mac. The reply comes back to the watch.
The author writes that no build from the source uploads anything. Record contents in Remote are end to end encrypted with keys in the user's iCloud Keychain. Only the text an alert shows stays outside that encryption, so iOS can display it. Agent HUD also carries its own MCP server. Through MCP, Claude Code can ask Codex to review a change and get the answer back inside Agent HUD.
Source
Primary source: jazzenchen/agent-hud-open on GitHub.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
