Supercharge the stack you already pay for.
Claude Code, Codex, and Cursor, together in one view. Pick a session from your Mac’s notch and get back to work.
One payment. $20 for your first Mac. $10 per additional Mac.
01Recognize the project.Not another mystery window.
02Pick up the thread.Return to your existing agent.
03Take help with you.In the notch or in a floating HUD.
A real app recording, from session card to agent window. Demo content is recorded. Your sessions stay in your tools.
01 / Sessions as context
Find the conversation you last worked on. Session cards keep the project, recent activity, and a way back to your agent together.
CPU, RAM, and disk as progress pills in the footer. Provider usage beside the companion.
CPU · RAM · Disk Footer progress pills. Usage meters sit beside the companion.
One place for the sessions and work across projects.
Claude, Cursor, and Codex sessions you already have, including Cursor cloud agents.
Ask the notch to summarize, then open the same agent with its history.
The expanded notch, captured from the app. Distinct sessions as context: pick up mid-project instead of re-explaining.
02 / Talk & Type
Hold Right Command to talk with your companion. Hold Right Option to dictate into the field you’re working in.
“Help me think
this through.”
Reach your voice companion from the notch. Ask the question while you still have the thought.
Here's what I'm thinking.
Let's keep the first version simple.
Focus your reply, document, or prompt. Phonon 2 turns your English speech into text on your Mac. Review the words and send them yourself.
Illustrated Talk and Type workflows. Example words, not a live recording.
On-device speech / Phonon 2
Phonon 2 is the default English speech engine for new Super Notch setups. Dictate into any focused field, or use the same local recognition when you Talk.
Fermion’s published English accuracy results. Lower word error is better.
| Speech model | Download | Word error |
|---|---|---|
| Phonon 2 Super Notch default | 164 MB | 5.21% |
| Parakeet TDT 0.6B v3 | 2,508 MB | 4.96% |
| Parakeet Redux | 178 MB | 5.69% |
| Whisper large-v3-turbo | 1,618 MB | 6.58% |
Source: Fermion Research, Table 1. Phonon 2 and Redux use Fermion’s evaluations; Parakeet and Whisper use the Open ASR Leaderboard rows cited there. These are model results, not an end-to-end Super Notch benchmark.
| Runtime | Times realtime |
|---|---|
| Phonon 2 · MLX | 174.0× |
| FluidAudio · Parakeet v3 · Core ML | 104.9× |
| FluidAudio · Parakeet Redux · Core ML | 27.8× |
| Moonshine tiny | 26.7× |
| whisper.cpp · large-v3-turbo · Metal | 17.0× |
| sherpa-onnx · Parakeet v3 · int8 | 16.5× |
Fermion Research, Table 3: 20 dictations totaling 797 seconds, one stream at a time, default runtime settings. Model loading is excluded. This measures transcription throughput, not the delay after releasing Super Notch’s hotkey.
English · Apple silicon · one-time model and runtime download. The runtime adds to the 164 MB model size. Existing engine choices are preserved; Whisper remains available for other languages or translation. Talk sends the transcript to your selected brain.
Voice features need macOS permissions. Point Super Notch at Grok or OpenAI in Settings for the best answers. Apple Intelligence can reply on supported Macs if you prefer on-device.
At a glance / Usage
Check Claude, Codex, Cursor, and Grok usage from the notch, alongside your Mac’s CPU, memory, and disk.
Usage comes from the accounts and tools you connect. Refresh from the notch and inspect each provider’s available limits and reset details.
See how usage worksAvailable windows and freshness vary by provider. Claude usage uses your signed-in Claude CLI.
The floating HUD
Ask about the webpage or app in front of you. Move the floating HUD over the window, then ask a question, get an explanation, or work through an idea. No copying into another chat.
Uses available window and page context. Content access depends on app support and permissions.
"Explain this paper."
00 / Your stack
Keep Claude, Cursor, Codex, and the rest. Super Notch finds the session and puts it on a card so you can get back in.
At a glance / Agent activity
Let your agents work in the background. Orbiting spheres in the notch keep their activity in sight, without switching apps just to check.
Larger spheres represent detected working sessions. Smaller spheres are ambient motion, not extra agents.
A brief burst marks a session no longer detected as working. A small signal to look back when you're ready.
Activity is inferred from supported sessions. A pop means a session is no longer detected as working, not that its result is verified; stale or disappearing sessions can also trigger it. Review the result in your agent's tool.
03 / Companion
Optional HUD companion that tracks your cursor without taking over the desktop. Sora and Hana stay optional.
Move your cursor. You've got company.
Chalk
04 / Same tools. Less friction.
05 / Studio
Release notes and the feature board. Ask for what you want next.
Compare / Type
Super Notch is not a monthly subscription. Type is on-device and $0/mo after the one-time license. Wispr Flow Pro is $15/user/mo.
| Measure | Wispr Flow Pro | Super Notch |
|---|---|---|
| Billing model | Monthly subscription | One-time Mac license. Not a monthly subscription. |
| Dictation cost / mo | $15/user/mo | $0/mo after you buy |
| License | Per-user Pro plan | $20 first Mac, $10 each additional Mac |
Contrast uses Wispr Flow Pro monthly pricing from wisprflow.ai/pricing. Wispr also lists annual billing. Super Notch Talk and Chat use your Grok or OpenAI plan; Type does not.
More comparisons: Wispr Flow alternatives · vs Superwhisper · vs NotchDeck.
Your notch. Put to work.
$20 for your first Mac. $10 for each additional Mac. Super Notch supercharges the stack you already pay for.
$10 for each additional Mac. USD. Apple silicon · macOS 14+.
Version 0.3.13 · signed and notarized.
Best with your Grok or OpenAI subscription. Type dictation uses no model tokens.
Pay here. Get a license key and a notarized download. Paste the key in Super Notch to activate this Mac.
No. Keep the tools you already use. Super Notch tracks and continues sessions across them.
No. $20 for your first Mac, $10 for each extra. One-time.
No. Type is on-device. After you buy, dictation is $0/mo. Talk and Chat use your Grok or OpenAI plan.
Yes. Move the floating HUD over the window and ask. What it can see depends on the app and permissions.
Yes. Buy a license, download the notarized app, and paste your key to activate this Mac.
No. It is another surface for Super Notch's conversation, with the same brain by default. You can give the HUD its own provider/model settings, and you can hide it entirely if you prefer the notch.
Place the HUD over the paper's browser window and ask, "Explain this paper." The explanation depends on the text available to Super Notch. If only the title or abstract is accessible, that is not evidence that it has read the full PDF; it should explain that limitation rather than invent missing content.
Super Notch is an AI agent session manager for Mac. It finds and resumes Claude Code, Codex, and Cursor sessions from the MacBook notch, with Talk from the notch and Type dictation into the focused field. It does not replace those coding tools.
Super Notch is a MacBook notch AI assistant: Talk from the notch, Type into the focused field, session cards, and a floating HUD. It runs on Apple silicon Macs, including MacBook Pro, on macOS 14 or newer. Cursor, Claude Code, and the rest stay the tools that write. Super Notch sits beside the agent.
Yes. Super Notch is a Mac notch AI companion. It lives in the notch and a floating HUD beside the agent you already run. Sessions, Talk, and Type sit next to Cursor, Claude Code, and Codex. Super Notch does not replace those tools.
Talk is the MacBook Pro notch voice assistant path. Hold Right Command, speak, and the answer uses Grok, OpenAI, or Apple Intelligence on a supported Mac. You need Apple silicon and macOS 14 or newer. Talk is the notch conversation, not a coding agent.
Yes. Focus that tool's field, hold Right Option, and Type writes the words there. You review and send. Type is local speech-to-text and spends no model tokens. Talk (Right Command) is the notch conversation, not a replacement for those tools.
Dictate into Cursor on Mac. Dictate into Claude Code on Mac.
No. Type (Right Option) is speech-to-text on your Mac. It pastes words into the field you focused. It does not call Grok or OpenAI, so it does not spend model tokens. Talk and Chat answers do use your chosen brain.
An Apple silicon Mac running macOS 14 or newer. Voice features need microphone and accessibility permissions. For the best answers, sign Super Notch into Grok or OpenAI in Settings. Apple Intelligence can answer on supported Macs if you want to stay on-device.
Handoffs go to visible agent sessions. You can follow the work in the receiving tool. Type enters text into the field you focus; you review and send it.
Opening Stripe Checkout…
Super Notch