Don't just dictate.
Let AI act.

Most voice apps clean up your speech. BrainJack does that — and goes further. Agent mode understands what you mean, remembers your session, and executes multi-step actions on any computer. Bring your own AI. Keep it private. No cloud required.

In beta · iOS on TestFlight · Android in progress · USB dongle available now

BrainJack — brain with USB connector
# Install on your computer (macOS / Linux):
curl -fsSL https://brainjack.ai/get | bash

# Windows (PowerShell):
irm https://raw.githubusercontent.com/scrappylabsai/brainjack-agent/main/install.ps1 | iex

Other apps clean your words. This one understands them.

Dictation apps give you text. BrainJack gives you two modes — clean transcription when you want it, and an AI agent that interprets intent when you need it. Switch with a single tap.

Dictation

Clean transcription

Say it, it types it. AI removes stutters, filler words, and false starts. You get clean text ready to send.

  • On-device speech recognition — nothing leaves your network
  • Manual mode (tap to speak) or Live mode (continuous listening)
  • Adjustable sensitivity — tune how long silence means "done"
  • Custom AI instructions — "formal tone", "coding style", "minimal"
  • Personal dictionary — teach it your jargon and proper nouns
Agent — no other voice app does this

Agentic dictation

Don't describe keystrokes. Describe what you want done. The AI interprets your intent, builds a plan, and executes — typing, pressing keys, triggering combos, waiting between steps.

  • Multi-step actions — type, key press, combos, delays, all in sequence
  • Session memory — remembers 10 turns of context per session
  • Verbal context — say "I'm filling out a vet form" and it adapts everything after
  • OS-aware — knows Mac vs Windows vs Linux, remaps shortcuts automatically
  • Growing toward RAG — persistent memory across sessions, your personal context

Two ways to connect. Or don't.

Install the service for WiFi control. Plug in a dongle for zero-setup USB. Or just use clipboard mode — no computer needed.

Install the Service

One command. Runs as a system service. Auto-starts on boot. Connects over your local network via WebSocket.

Works on: Windows (SendInput) · macOS (osascript) · Linux X11 (xdotool) · Wayland (ydotool)
+
Security: HMAC token auth · TLS optional · rate limiting · audit log (never logs keystrokes)

Plug In the Dongle

No other voice app can do this. An ESP32-S3 USB device that appears as a standard keyboard. No drivers, no software, no admin rights on the target machine. Ever.

Works with: BIOS · login screens · air-gapped machines · anything with a USB port since 1995
or

Clipboard mode

No connection at all. Speak, AI generates, copied to your clipboard. Paste wherever you want.

In the beta right now.

Everything here is in the current TestFlight build. Working, but we're still testing. That's why we need you.

Live Mode
Continuous listening with adjustable sensitivity. Speak naturally — the engine keeps listening after each result. No re-tapping. Sensitivity slider controls silence timeout from 0.8s to 3.0s.
Command Sheets
One-tap macros for Vim, tmux, Emacs, VS Code, and Shell. Tool-specific palettes with 15–50 buttons each. OS-aware — Cmd on Mac becomes Ctrl on Windows automatically.
Multi-Device
Save multiple computers. Switch between your work Mac, home Linux box, and a cloud VM. Each device remembers its OS type, connection method, and auth token.
QR Setup
Scan a QR code from the service running on your computer. IP, port, and auth token — configured in one scan. Manual entry also available.
Kill Switch
Red button. Sends emergency stop. Clears any stuck HID injection immediately. Because sometimes you need to pull the brake.
Offline Queueing
Lose connection mid-dictation? Commands queue locally and replay when your device reconnects. Nothing lost.

What you won't find anywhere else.

Voice dictation is a crowded market. Most apps clean your speech into text. Here's what they don't do.

Agentic AI
Other apps transcribe. BrainJack interprets intent and acts — multi-step keyboard actions, session memory, verbal context that accumulates. Say "I'm in a web form" once and every action after adapts. No other voice app has this.
Bring your own AI
Other apps lock you into their cloud. BrainJack lets you plug in your own ASR backend, your own LLM, your own API keys — any OpenAI-compatible endpoint. Use our services or don't. Your choice.
Private by architecture
Not "we promise we won't look at your data." There is no cloud to breach. On-device ASR. Local network only. Keystroke content is never logged. Other apps are SOC 2 certified — we don't need to be, because your voice never leaves your network.
USB HID hardware
A physical dongle that types into BIOS, login screens, air-gapped machines — anything with USB. No software on the target. No other voice product in the market offers this. It's a different category.
Real Linux support
Most voice apps are Mac-first, Windows-maybe, Linux-never. BrainJack supports X11 (xdotool) and Wayland (ydotool) as first-class targets. Because developers use Linux.
Developer tools
Pre-built command sheets for Vim, tmux, Emacs, VS Code, and Shell. One-tap macros. OS-aware shortcuts. This isn't a productivity app for note-takers — it's a power tool for people who live in the terminal.

Real scenarios. Not hypotheticals.

Developer on the couch

SSH into servers by voice. One-tap Vim macros. Dictate commit messages. Command sheets built for people who live in the terminal.

Multi-machine operator

Switch between your work Mac, home Linux box, and a cloud VM — all from one phone. Each device remembers its OS and macros.

Filling out forms

"Write a letter confirming my dog's rabies shot on March 15th." Agent mode generates the full letter. Review on your phone, paste into the web form.

Hardware hackers

Flash an ESP32-S3 or Flipper Zero. Open firmware on GitHub. Build a voice-controlled anything for the cost of a coffee.

The agent gets smarter.

BrainJack isn't a dictation app with AI bolted on. It's a platform. The intelligence grows.

Now

Session memory

The agent remembers your last 10 turns. Verbal context accumulates — say "I'm in a web form" once, and every action after adapts. OS-aware actions, multi-step sequences.

Next

RAG + persistent context

Memory that survives sessions. The agent knows your preferences, your tools, your workflows. Retrieval-augmented generation over your personal context.

Then

Autonomous actions

Multi-app workflows. Voice-triggered automations. The agent doesn't just type — it navigates, fills, submits, and verifies. Your voice becomes a control surface for everything.

Private by architecture.

Not "privacy-friendly." Private. All processing stays on your network. There is no cloud to breach.

On-device ASR

Speech recognition runs on your phone. Audio never leaves the device. Bring your own ASR backend if you want — it's your choice.

Token auth

HMAC-SHA256 constant-time comparison. Generated during install. Shared via QR code.

Audit logging

Every connection, auth attempt, and command type is logged. Keystroke content is never recorded.

TLS + rate limiting

Optional TLS encryption. Per-IP rate limiting prevents abuse. Reverse proxy mode for production setups.

Your keys. Your infra. Our app.

The service is source-available and free for personal use. The app is what you buy. The dongle is a separate physical product. During beta, everything is free to test.

You're not locked in. The app lets you bring your own ASR backend, your own LLM, your own API keys. Point it at your own infrastructure and it just works. ScrappyLabs services are there as a convenience — not a requirement. You buy the app because it's the best voice-to-action client, not because we trapped you.

Source-available

BrainJack Service

Free for personal use

The backend that receives commands and injects keystrokes. Install it on any computer. Source on GitHub under BSL 1.1 — free for non-commercial use, converts to Apache 2.0 in 2030.

  • macOS, Windows, Linux (X11 + Wayland)
  • One-command install, auto-starts on boot
  • Token auth, TLS, audit logging
  • 683 lines of Python. No bloat.
Hardware

USB Dongle

Available now

Pre-flashed ESP32-S3 USB HID controller. Plug it into any machine — appears as a standard keyboard. Nothing to install on the target. Works with the app over WiFi.

  • BIOS, login screens, air-gapped machines
  • No drivers, no software, no admin rights
  • Open firmware on GitHub
  • Works with anything USB since 1995

Email for pricing and ordering

# Install the service (free for personal use):
curl -fsSL https://brainjack.ai/get | bash

# Get the iOS app (free during beta):
testflight.apple.com/join/GTUVUYyr

We're in beta. Come break it.

Install the service, grab the app, and try everything. All features unlocked during beta. Bring your own AI backends or use ours. We need real users testing real workflows.

Install from GitHub Press & partnerships