AgentPeek - Your coding agents, in the Mac notch
by•
You're running more coding agents than ever, but you can't keep up with them. That's where AgentPeek comes in. It pulls every session up into your Mac notch, live. Glance up, approve a prompt, watch token usage and manage the entire flow without pausing your YouTube video. All local, all yours.
Replies
AgentPeek
Hey Product Hunt 👋 I'm Bren, the solo dev behind AgentPeek.
I basically live in Claude Code and Codex now. The problem: too much time spent babysitting them. Manually checking tabs, usage, servers, whether anything's actually finished, all while I'm just trying to watch some YouTube.
IT DOESN'T HAVE TO BE LIKE THIS!
So I built AgentPeek. It puts every agent's live status right in your MacBook notch. One glance tells you what's running, what needs you, and what just finished. Lovely!
The part I'm most excited about is how little it asks of you:
- Live status in the notch. Running, waiting on you, or done, at a glance.
- Approve or deny permission prompts straight from the notch, without leaving the app you're in.
- Watch each tool call as it happens (reads, edits, commands), each with its own glyph.
- Zero setup. It reads the hooks Claude Code and Codex already fire, so you just open it and it works.
And that's just the start. There are dozens of small quality-of-life touches baked in too, the kind of thing I added because it bugged me as a dev and now I can't work without them.
I built this for my own workflow first, so I'd genuinely love your feedback.
QUESTION: what do you need out of this app?
I'm shipping constantly and focused on practical, time-saving stuff, so tell me what would actually help.
Thanks for checking it out 🙏
p.s. I post what I'm building on X at @brenhubr if you want to follow along and connect.
Radar By Paved
@brenhubr @brenhuber this looks really useful (unlike a lot of notch apps). random question: the hero video on the website is clean. curious if you used an ai videogen platform to make it or was it a custom edit? looking to create something similar.
AgentPeek
@brenhubr @dannygreer Thank you, I used Screen Studio for it!
AgentPeek
@priyatharshini_c Thank you! It queues all pending approvals, you can answer one after another.
@brenhuber Queueing makes sense, keeps it simple and avoids decision fatigue from seeing everything at once. Thanks for the quick answer!
The notch is a great spot for this, I usually have three or four Claude Code sessions going and lose track of which one is blocked waiting on me. Does it show which repo each session belongs to when you have a bunch running? congrats on shipping
AgentPeek
@i_sanjay_gautam Absolutely, shows the name of it in the sessions list! :)
The status I actually want at a glance is whether a session is blocked on me or just slow, because a long tool call and a session sitting on a permission prompt look identical from the outside. Running several agents at once, the time I lose isn't checking usage, it's tabbing into a session to find it's been waiting ten minutes for a y/n. If the notch can flag the blocked-on-human state distinct from busy, that alone earns the space. Does it read the actual pending prompt, or infer from idle time?
AgentPeek
@dipankar_sarkar It reads the actual prompt, not idle time. The CLI fires a permission hook the moment a tool call needs approval, so "blocked on you" is its own state, not a session that just went quiet. The notch shows what it's waiting on and lets you approve or deny right there.
Pulling sessions into the notch solves the context switching problem that comes with running multiple agents in parallel. The local-only data model is the right call for codebases you can't risk leaking. We've felt the friction of constantly alt-tabbing to check which agent is waiting on approval. How do you hook into Claude Code sessions: process output, local socket, or something else?
AgentPeek
@anand_thakkar1 Native hooks! Both agents support hooks, and they are wired into .claude/.code settings.json!
The local-only session view is the important part here. Once coding agents touch real repos, the hard problem becomes knowing what is running, what is blocked on a human, and what can safely be approved without turning the whole workflow into a terminal babysitting loop.
AgentPeek
@krekeltronics Thank you! Always in the loop :)
running 3-4 Claude Code sessions at once is where this starts making real sense. the thing i keep bumping into though - does it show aggregate token spend across all sessions? each session looks fine individually but cumulative burn across 4 parallel agents adds up way faster than expected. a total daily spend view would make the budget awareness actually useful
AgentPeek
@omri_ben_shoham1 I will add this in the token usage view, thank you :)
Bren, the notch sits empty most of the day, so quietly putting something useful up there seems very clever. I like that everything it watches stays on my own computer ! Thanks
AgentPeek
@robin_de_lacroix Thank you! All local :)
I didn't know what mac notch is, and now i downloaded it. any plans to make it hermes compatible? I know session mngment is less of an issue there but some of us have more than two harnesses.
Congrats on the launch. Trying it out right now!
Also very cool landing.
AgentPeek
@arjinroni I will look into more compatibility, Hermes has grown so fast! Thank you :)
Using the permission hook as the signal is the right call, idle-time heuristics would've flagged half my long tool calls as stuck. The one state that still looks identical from the notch is a tool call genuinely running for twelve minutes versus one that's quietly wedged, since neither trips the approval hook. Do you surface elapsed-time-in-current-tool, or just the blocked vs running split?
AgentPeek
@dipankar_sarkar Yep, you can expand the session and see the time gap between each tool :)