AI assistant
Give Claude the long version of what you want
Claude is built to take a lot of input. Most people give it a fraction of what they have, because the rest lives in their head and typing it out is a chore.
Free tier runs on-device. macOS 14+ on Apple Silicon or Intel.
Workflow
What to use voice for in Claude
- Long briefs where the requirements, the constraints and the things to avoid all need saying
- Feeding a spec you are holding verbally before it exists as a document anywhere
- Asking for a rewrite with the reasons attached, not just "make it better"
- Adding standing context to a Project so you stop re-explaining how your team works
- Reacting to an artifact out loud: what is wrong with it, which parts to keep, what to try instead
Friction
Where typing slows Claude down
- A three-paragraph prompt takes several minutes to type and about forty seconds to say
- Nuance gets cut first, and nuance is exactly what a long-context model can use
- Rewrite requests come out vague because the specific objection is harder to articulate in text than in speech
- Project instructions are written once, in a hurry, and never expanded
Examples
Things you would actually say
Spoken out loud, at normal pace. The words below are the input, not the output — Hyprcore types them where your cursor already is.
“Here is what I need. We are adding usage-based billing on top of the existing seat plans. Customers keep their seats but get charged for events above their included allowance, which is ten thousand a month on starter and a hundred thousand on team. Invoices go out on the first, usage resets at midnight UTC on the first, and any overage in the final three days of a cycle rolls into the next invoice rather than the current one. Do not touch the trial logic. Write me the data model and tell me where you think this breaks.”
“Rewrite the second section for a finance director rather than an engineer. Keep every number exactly as it is, drop the words latency and throughput, and get it under two hundred words. The current version buries the cost saving in the last sentence, which is the only part they will care about.”
“For the whole of this project, assume the following. We ship on the second Tuesday of each month. The public API is on version three, and the legacy webhooks endpoint was deprecated in January and disappears in October. Every customer on the starter plan is capped at ten thousand events a day. When you write examples, use British spelling and the currency in pounds.”
The ask that deserves three paragraphs gets one line
The gap is between what you know and what you are willing to type. You could explain the whole billing edge case in forty seconds to a colleague standing next to you, and instead you type "help me design usage-based billing" and spend the next four turns feeding back the constraints one at a time as Claude guesses wrong about them.
Dictation collapses that. Speaking a long brief costs roughly what saying it out loud costs, so the incentive to trim disappears, and the version Claude receives is closer to the version in your head. This is the specific reason voice suits Claude more than it suits a search box: the input side is where the leverage is.
The clean-up matters too, because spoken briefs ramble. Select what you dictated, say "Hyper, organise this into requirements and constraints," and Action Mode rewrites the selection rather than adding to it. That runs on the text you have selected in any app, so it works the same in the Claude desktop app, in a browser tab, or in the note where you drafted the brief first.
Frequently asked questions
Can I dictate several paragraphs into one prompt without it breaking up?
Yes, and long input is the case this is most useful for. Text arrives in phrases as you speak rather than in one block at the end, so a long brief builds up in front of you and you can see whether you have covered everything before you stop.
Does it work inside Projects and in artifact editing?
It works in any field you can put a cursor in, which covers the main composer, project instructions, project knowledge and the boxes you type into when reacting to an artifact. Hyprcore does not know it is in Claude, and it cannot edit an artifact itself; it only types where you are already typing.
Will it transcribe technical terms in a spec correctly?
Not reliably out of the box, which is what custom vocabulary is for. Add your service names, internal acronyms and product codenames, and the bias applies to both the on-device and the cloud transcription paths, so the terms you use twenty times a day stop being guessed at phonetically.
Can I use this to dictate to Claude on my phone?
No. Hyprcore runs only on macOS, on Apple Silicon. There is no iOS app and no browser version, so dictating into Claude on a phone means using whatever your phone provides.
Dictate somewhere else
Last reviewed July 25, 2026.
Try it
Talk into Claude.
Free tier transcribes on-device on your Mac — no credit card required.
Native macOS · On-device dictation · No credit card