still writing ai slop?
that's embarrassing 😬

your readers can tell. hyv catches it before you ship. get started for free!

npm i -g @holdyourvoice/hyv
works across all your mac/windows apps.
01 · voice score
34/100

6 issues · 3 red · 3 yellow · local scan

02 · flagged
03 · command
$ hyv

checks draft.md offline — zero api calls

04 · output
// stop patching ai slop

ai slop is real.
your voice is not.

llms generate the same writing for everyone. your vocabulary, your patterns, your taste. ai writing should flow like you think.

without hyv score 31

in today's fast-paced digital landscape, it's worth noting that leveraging robust frameworks can streamline your workflow. moreover, this holistic approach fosters seamless collaboration across teams.

delve leverage moreover holistic it's worth noting
with hyv score 89

last march i shipped a dashboard for a client. three months of work. she called on a thursday — nobody had logged in since launch. turns out her customers were just texting her questions directly.

specific moment your cadence short sentences no hedge words
250+
patterns caught locally
100%
offline. no upload.
9+
apps, one install
// why not just a prompt?

a prompt template can't learn.
hyv does.

static system prompt

works once. drifts fast.

  • you paste "write in my voice" into a system prompt
  • model matches your tone for a day or two
  • then defaults back to generic ai cadence
  • no memory when your vocabulary evolves
wk 1
72%
wk 4
38%
wk 8
19%
voice match fades — you're re-prompting every week.
hyv voice profile

compounds every edit.

  • profile built from your actual writing samples
  • every accept/reject teaches your never-list
  • hyv reinforce sharpens patterns you keep
  • one profile works in cursor, claude, codex, windsurf
wk 1
68%
wk 4
84%
wk 8
93%
voice match climbs — you explain less, ship more.
// three steps

install once. use everywhere.

hyv works inside the ai apps you already use — claude, chatgpt, codex, cursor. it never calls its own llm. your model does the work, you own your tokens.

install globally

one npm install. hyv works in every terminal and ai agent on your machine. (requires node.js.)

npm i -g @holdyourvoice/hyv@latest

scan any draft

run the live demo or scan a draft. hyv flags generic buzzwords, fake storytelling, and robotic cadence. free, offline.

hyv welcome
hyv scan draft.md

build your voice profile

when you're ready for rewrites that compound, sign in and create a profile from your writing. one profile, every ai app.

hyv init
hyv new my-voice

agents: hyv mcp --setup for cursor & claude. prefer the browser? 30+ free web tools — no install.

// free cli

scan locally. free forever.

the npm package works offline for scan, fix, check, score, and diff. no subscription, no api calls. your text never leaves your machine.

terminal commands

mac, linux, or inside cursor/claude via mcp. all offline.

web tools (30+)

same engine in the browser — no install, no account.

paid unlocks profiles, learning, and hybrid analysis — hyv plan --free shows the full matrix.

hyv scan draft.md
// what happens after you install

paste a draft. see the ai. fix it.

1
step one

drop in a draft

save your chatgpt or claude output to a file — or paste straight into the terminal. any format works.

newsletter-draft.md

it's worth noting that robust systems are the cornerstone of sustainable growth. moreover, leveraging automation can streamline your operations and enhance productivity across the board.

2
step two

hyv shows you the ai

local scan flags buzzwords, fake storytelling, and robotic cadence — line by line, before anything ships.

hyv scan newsletter-draft.md
L1 ai_vocab robust · leverage · streamline
L1 connector moreover
L1 hedge it's worth noting
score 34 / 100 · 6 issues
3
step three

fix it in your voice

hyv fix swaps what it can locally. your agent rewrites the rest using your profile. accept, reject, reinforce.

after hyv fix + rewrite

last march i shipped a dashboard for a client. three months of work. she called on a thursday — nobody had logged in since launch. turns out her customers were just texting her questions directly.

score 91 / 100 · clean
// what hyv catches

your readers can tell.
hyv shows you why.

paste any ai draft. hyv scans it instantly and shows you every line that doesn't sound like you.

hyv scan blog-post.md
blog-post.md: 0 issue(s)
L3ai_vocab_density "delve" "robust"
L7founder_cadence "here's the thing"
L12formulaic_connector "moreover"
L15inflated_verbs "supercharge your growth"
L0low_burstiness every sentence is 15-20 words
L0voice_no_storytelling no personal moments in 800 words

250+ patterns across vocabulary, rhythm, tone, and structure.

claude claude code chatgpt codex cursor windsurf antigravity opencode command code
// community

founders, builders, and writers use it daily.

"i used to spend 30 minutes per post untangling chatgpt's voice. now i run one command, accept the rewrites that fit, and ship. the learning loop is the part i didn't expect to need."
gs thina
gs thinacommunity builder
"our team writes 40 blog posts a month. before hyv every one of them sounded like the same template. after two months the profile knows our style better than our new hires do. new drafts land in our voice, not chatgpt's."
hameed raha
hameed rahab2b saas founder
// automations

write once. enforce everywhere.

hyv runs in ci/cd, cloudflare, n8n, anywhere code runs. scan every draft before it ships. automated. no manual steps.

ci/cd

github actions

block merges when ai patterns slip into blog drafts or marketing copy.

pull request check
hyv scan content/ --fail-on-hit
exits 2 on any hit — fails the build
local

watch while you write

re-scan on every save while you edit in cursor, vscode, or any editor.

live file watch
hyv watch draft.md
score updates each time you save
batch

scan a whole folder

run across every markdown file in a repo — blogs, docs, changelogs, all at once.

glob scan
hyv batch "**/*.md" --fail-on-hit
catches drift across every file

works on any mac. open terminal once, paste one command, done. your voice profile travels with you across every ai tool you use.

// pricing

first month for a dollar. then $9/mo.

your voice profile compounds in value the longer you use it. worth paying for.

multiple
$29/mo
for teams and agencies

multiple profiles, shared workspace, same voice gate everywhere.

  • unlimited voice profiles
  • team sharing
  • unlimited scans and rewrites
  • profile learning loop
  • cli + dashboard access
  • automation support
get started →

people can tell. every single time.

your reader knows in seconds when something was written by ai. they don't need a detector — they just feel it. hyv keeps your writing sounding like you, not like a template.

npm i -g @holdyourvoice/hyv@latest

free local scan forever · profiles from $1 first month

// faq

questions, answered.

does hyv call an llm?

no. hyv tells YOUR ai app how to rewrite — you use your own chatgpt or claude account. no extra cost, no extra subscriptions. hyv just makes sure the output sounds like you.

what ai apps does this work with?

any ai app that runs on your computer: claude code, chatgpt desktop, codex, cursor, windsurf, and more. it also works from the terminal if you prefer. basically — if you can paste text into it, hyv works with it.

how does the voice profile work?

built from your writing — vocabulary fingerprint, sentence rhythm, emotional tone, signature phrases, banned patterns. captures how you actually write, not how you describe yourself. improves every time you accept or reject a rewrite.

is my writing private?

scans happen locally. your text never leaves your machine. only the voice profile and learning signals are stored server-side.

what if i don't have an ai app yet?

download claude for desktop (free) — it's the easiest place to start. install hyv, build your profile, and paste your drafts directly into claude. no terminal needed after setup.

can i cancel anytime?

yes. cancel from the dashboard, the cli, or reply to any billing email. your voice profiles stay yours and you can reactivate anytime.

try for $1 →