cd /news/ai-products/yap-free-offline-voice-dictation-for… · home topics ai-products article
[ARTICLE · art-42232] src=github.com ↗ pub= topic=ai-products verified=true sentiment=↑ positive

Yap – free offline voice dictation for Mac/Windows/Linux (Wispr Flow alt)

Yap, a free and open-source offline voice dictation tool for Mac, Windows, and Linux, has been released as an alternative to Wispr Flow and SuperWhisper. It runs locally using Whisper for transcription, requires no account or subscription, and does not send audio or screenshots to the cloud unless the user opts in. The tool is MIT-licensed and supports custom endpoints for cloud engines and LLM cleanup.

read11 min views1 publishedJun 28, 2026
Yap – free offline voice dictation for Mac/Windows/Linux (Wispr Flow alt)
Image: source

Hold a key, talk, and what you said lands at the cursor: your editor, browser, a chat box, the terminal, wherever you're typing. It's a free, open-source alternative to Wispr Flow and SuperWhisper, and it runs offline by default.

What you get:

It stays private. Transcription runs locally with Whisper, so there's no account, no cloud, no telemetry, and nothing screenshots your screen. Audio only leaves your machine if you deliberately switch to a cloud engine.No word caps, no subscription. Wispr's free tier stops at 2,000 words a week and then runs $12–15/mo. yap is MIT-licensed, with no meter and nothing to upsell.Same on every OS. macOS, Windows and Linux share one config file.Quick. Faster than real time on CPU with thebase

model, near-instant with a GPU or a cloud key.Your keys, your endpoints. Point it at OpenAI, Groq, or a Whisper server you host yourself. There's an optional LLM cleanup pass too, against any OpenAI-compatible API.

  you: (hold Right Option ⌥)  "send him the q3 numbers by friday"
  yap: Send him the Q3 numbers by Friday.        ← typed at your cursor

Grab it, unzip, open. That's the whole thing:

| |---|

⬇ Windows

⬇ Linux

macOS:not Apple-signed yet, so the first timeright-click the app → Open(not double-click), then grant Microphone + Accessibility once. Comfortable with a terminal? See[Install]below.

| Wispr Flow | yap | | |---|---|---| | Price | $12–15/mo (free tier: 2k words/wk) | Free, unlimited | | Offline mode | ❌ cloud only | ✅ local Whisper by default | | Sends audio to a server | Always | Only if you opt into a cloud engine | | Screenshots active window | Yes (for "context") | Never | | Open source | ❌ | ✅ MIT | | Platforms | Mac/Win/iOS/Android | Mac/Win/Linux |

Requires Python 3.9–3.13 (3.14 is too new — some native deps don't have wheels for it yet). The local engine downloads a small Whisper model on first use. Clone, run the installer, go:

git clone https://github.com/AkuchiS/yap.git
cd yap
./install.sh
yap run

./install.sh

sets up an isolated pipx environment so nothing pollutes your system Python. Then hold Right Option, speak, and let go; your words land at the cursor.

Prefer a one-liner?

pipx install "yap-dictation[full] @ git+https://github.com/AkuchiS/yap"

does the same thing. The[full]

extra pulls in the macOS menu-bar bits (rumps

); plaingit+…

installs the core CLI only, soyap app

won't start without it.

macOS— grant your terminal** Accessibilityand Microphone**permission:System Settings → Privacy & Security → Accessibility / Microphone. Without Accessibility, the paste keystroke is silently ignored.Linux (X11)— works out of the box. For clipboard fallback install one ofxclip

/xsel

. OnWayland, installwl-clipboard

; some compositors restrict synthetic keystrokes — use--engine local

withinject.method = "type"

if paste doesn't land.Windows— no extra steps; run from a normal terminal.

yap run                       # start the daemon; hold your hotkey and talk
yap app                       # tray app: menu-bar (macOS) / system tray (Win/Linux)
yap transcribe meeting.m4a    # one-shot: transcribe a file, print the text
yap vocab add PostgreSQL      # teach it your words (see below)
yap hardware                  # show your specs + the model it'll auto-pick
yap doctor                    # diagnose permissions / hotkey / mic / clipboard
yap devices                   # list microphones
yap config show               # print effective config
yap license                   # show your install date + early-adopter code
yap update                    # check GitHub for a newer release

The default model is "auto"

— yap detects your CPU/RAM/chip and picks a size that stays responsive: tiny.en

on very old/light machines, base.en

on a 2019-era laptop, small.en

on modern hardware, GPU-accelerated if you have CUDA. Run yap hardware

to see the pick, or pin one with yap config set local.model '"small"'

.

yap uses your system default mic, which is usually fine. Where it bites is docking a laptop to a monitor. Plenty of monitors, USB-C docks and webcams have their own mic built in, and the OS tends to make that the default the moment you plug in, so you end up recording from a mic across the room (or one that doesn't really work). It's not a mac quirk; Windows and Linux docks do the same thing.

The fix is to pin the mic you actually want:

yap devices                                       # list mics, and show which one yap picks
yap config set audio.device '"Studio Display"'    # match by name (substring, case-insensitive)

If you dock and undock a lot, give it a list instead of one name. yap re-checks on every keypress and uses the first one that's actually plugged in, so closing the lid (built-in mic gone) just falls through to the next:

yap config set audio.device '["MacBook Pro Microphone", "Studio Display"]'
yap config set audio.device '["Built-in", "Dell Monitor", "USB"]'   # Windows/Linux style

yap devices

prints a yap will use: …

line so you can see what it picked.

Some external mics only run at 48 kHz. yap records at whatever rate the mic offers and resamples to 16 kHz itself, so those are fine. If the recording light comes on but nothing gets typed, run yap run --debug

in a terminal and it'll say why: it couldn't open the mic, no audio arrived, or what it heard was silent.

yap only holds the mic while you hold the hotkey, so it doesn't fight your other tools. If you run your own always-listening assistant, point the integration hooks at it so it s while you dictate and resumes after:

yap config set integration.on_record_start '"myassistant "'
yap config set integration.on_record_stop  '"myassistant resume"'

Each hook runs with context in its environment, so the handoff can be smart about the OS and which app you're dictating into:

var value
YAP_EVENT
start or stop
YAP_OS
darwin / win32 / linux
YAP_ACTIVE_APP
the frontmost app (e.g. Slack , Code , Terminal )

For example, don't your assistant when you're dictating into it:

yap config set integration.on_record_start \
  '"[ \"$YAP_ACTIVE_APP\" = \"MyAssistant\" ] || myassistant "'
yap config set integration.on_record_stop \
  '"[ \"$YAP_ACTIVE_APP\" = \"MyAssistant\" ] || myassistant resume"'

Same context is written to integration.state_file

as {"active": bool, "os": "...", "active_app": "..."}

if you'd rather poll.

By default yap is push-to-talk on a single key (like Wispr's "hold to dictate"): hold Right Option ⌥ on macOS (Right Ctrl elsewhere), speak, and release to transcribe.

Dictating into different apps? Give each one its own settings. yap merges the matching app's overrides over your base config for that utterance, matched by the frontmost app's name — perfect for app-specific jargon, or turning the LLM cleanup on in chat but off in your terminal:

yap config set 'app_profiles.Slack.cleanup.enabled' true
yap config set 'app_profiles.Terminal.vocabulary' '["kubectl","stderr","grep"]'
yap config set 'app_profiles.Code.local.language' '"en"'

Matching is exact first, then a case-insensitive substring (so Slack

also matches an app reported as Slack — chat

). Anything you don't override falls back to base.

yap config set hotkey.combo '"<alt_r>"'    # Right Option ⌥ (macOS default)
yap config set hotkey.combo '"<cmd_r>"'    # Right Command
yap config set hotkey.combo '"<f9>"'       # a function key
yap config set hotkey.combo '"<ctrl>+<alt>"'   # a two-key combo
yap config set hotkey.mode  '"toggle"'     # press to start, press to stop

Why not the Fn / 🌐 key?It's the obvious one-finger choice, but on macOS the Fn key emits no real keypress — only a hidden hardware flag — so the cross-platform input library can't see it. Right Option gives you the same one-key feel and works reliably. (A native Fn backend is on the roadmap.)

Like Wispr, yap can learn the names and jargon you use so they come out right instead of being guessed at ("PostgreSQL", not "post grey sequel"):

yap vocab add PostgreSQL Anthropic Kubernetes # bias recognition toward these
yap vocab fix "github" "GitHub"               # always rewrite a misheard word
yap vocab list

vocab add

feeds Whisper a glossary hint (helps it spell unfamiliar words); vocab fix

is a guaranteed find/replace for words it consistently mangles.

By default yap watches what you dictate and learns the proper nouns, jargon, and acronyms you use repeatedly — adding them to your glossary automatically (capped, persisted, and never one-offs). So names like PostgreSQL

or Kubernetes

start coming out right after you've used them a few times, with no effort.

yap vocab learned          # see what it's picked up
yap vocab forget Foo       # drop something it learned by mistake

Tune or disable it in config under learning

(enabled

, min_count

, max_words

). It only ever learns repeated words and is capped, so it can't run away or hurt accuracy.

By default yap prints just ● listening…

and the final ✓ "transcript"

. Want silence (e.g. running it as a background app)? yap run --quiet

. Debugging? yap run --debug

narrates every stage.

Config is a JSON file (yap config path

to find it). Secrets are never stored here — API keys are read from environment variables. Common tweaks:

export YAP_API_KEY=gsk_...                 # e.g. a Groq key
yap config set engine '"cloud"'

export OPENROUTER_API_KEY=sk-or-...
yap config set cleanup.enabled true

yap config set audio.device 3

yap config set inject.method '"type"'      # if clipboard paste misbehaves

Engines:

engine | What runs | Cost | Privacy | |---|---|---|---| local (default) | faster-whisper on your CPU/GPU | free | audio never leaves the machine | cloud | any OpenAI-compatible /audio/transcriptions | your key | audio sent to that endpoint |

Local model sizes (set local.model

): auto

(default — adapts to your machine), tiny.en

, base.en

, small

, medium

, large-v3

. Bigger = more accurate, slower, more RAM.

For a proper app that shows yap + your icon in macOS permission dialogs (not "Python") and needs no separate Python install, freeze it with PyInstaller. The app bundles its own interpreter + the Whisper stack.

yap icon ~/Downloads/yap-icon.png        # (optional) your icon first

./packaging/build_macos.sh

./packaging/build_linux.sh

powershell -ExecutionPolicy Bypass -File .\packaging\build_windows.ps1

Tips: set YAP_BUILD_PY=python3.12

to freeze with a specific (mature) Python; YAP_MENUBAR_ONLY=1

hides the macOS Dock icon. After building on macOS, open the app and grant yap Microphone + Accessibility + Input Monitoring — now shown under the yap name and icon. Verify any build with yap selftest

.

If you just want a Dock launcher fast and don't mind that permissions show under "Python", yap bundle --login

makes a lightweight ~/Applications/yap.app

that calls your installed yap. Good for personal use; the frozen build above is the one to ship.

Linuxinstall/yap.service

(a user systemd unit), or addyap run

to your desktop's autostart.Windows— a shortcut toyap run

in the Startup folder.

See install/ for ready-made helpers.

yap needs two macOS permissions: Input Monitoring (to see your hotkey) and Accessibility (to type at your cursor). macOS doesn't always prompt — so the reliable way is to add yap by hand:

  • System Settings → Privacy & Security → Input Monitoring→ click→ press+

⌘⇧G, enter/Applications/Yap.app

,Add, toggle it** on**. - Do the same under Accessibility. Quit yap fully and reopen it— grants only take effect on relaunch.

Running the CLI (yap run

) from a terminal instead of the app? Grant your terminal app those two permissions rather than Yap.app.

Rebuilt the app and it stopped working? An unsigned app's identity changes every build, so old grants go stale and show "on" but don't apply. Reset and re-add: tccutil reset All com.yap.dictation

, then repeat the steps above. (A single downloaded build doesn't have this problem — it's only an issue when you rebuild repeatedly.)

With the default local

engine, yap is 100% offline — disconnect your network and it still works. It records audio only while you hold/toggle the hotkey, keeps nothing on disk, and phones no one home. Cloud engines and the optional cleanup pass are strictly opt-in and only ever talk to the endpoint you configure.

yap stamps the date of its first run on your machine — a tiny local install.json

, kept on-device and never uploaded. There's no paywall today. But if a paid tier ever appears, everyone who installed before that day keeps the free version: run yap license

to see your install date and a portable code that proves it.

yap is free and open source — built by a retired veteran, in his spare time, because dictation shouldn't be locked behind a subscription. If it saves you time, a small donation helps cover real costs (like Apple's $99/yr signing fee for warning-free Mac builds) and keeps it moving. Totally optional. yap is free today — and early adopters stay free for good: if a paid tier ever arrives, everyone who installed before it keeps the free version (see Early adopters).

MIT — see LICENSE. Built on faster-whisper / OpenAI Whisper.

── more in #ai-products 4 stories · sorted by recency
── more on @yap 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/yap-free-offline-voi…] indexed:0 read:11min 2026-06-28 ·