Just Speak.
It Acts.

The action-first voice agent that executes, not just transcribes. Draft emails, run commands, rewrite text—all by voice. Zero context switching.

Platform independent and fast
Download for macOS
See What It Does ↓
Scroll to explore
⚡ Sub-100ms Latency · 🛡️ Fully Open Source · 🧠 On-Device AI Engine · 🔌 MCP Protocol Support · 🎙 Voice-to-Action Pipeline · ⚡ Sub-100ms Latency · 🛡️ Fully Open Source · 🧠 On-Device AI Engine · 🔌 MCP Protocol Support · 🎙 Voice-to-Action Pipeline ·
Why Speed Matters

Sub-100ms Latency.
Locally Powered.

Powered by Whisper & Moonshine running entirely on your machine. Your voice becomes text before your finger lifts off the hotkey. No server round-trips. No waiting. No excuses.

<87 ms* Average transcription latency on Apple Silicon M4
⚡

On-Device AI Engine

HeyDict is local-first by choice, with cloud fallback to guarantee speed on any device. Your voice becomes text instantly, anywhere.

🧠

Hybrid Semantic Routing

A local embedding model routes common commands instantly. Only complex, context-heavy tasks fall back to Cloud LLMs—keeping 80% of operations offline.

💉

Universal Text Injection

Injects transcribed or AI-generated text directly into the active window via clipboard and accessibility APIs. Works across every app, every OS.

🔌

MCP Protocol Support

Connect to any MCP-compatible tool server. Your voice commands extend into your entire toolchain—databases, APIs, internal services.

Action-First Workflows

It Doesn't Chat.
It Executes.

HeyDict is not a chatbot. It's an always-on overlay that turns your voice into direct action across every application on your desktop.

Terminal Fix

From Terminal Error
to Fixed Build

ERROR: cannot find module 'react-dom/client'
at /app/src/index.tsx:3:1
⌨️ HeyDict detected error. Fixing...
$ npm install react-dom@latest
✓ Build succeeded in 2.1s

HeyDict sees your terminal errors in real-time, understands the context, and can execute the fix—all before you finish reading the stack trace.

🔧 Auto-Fix Mode
Slack Reply

The 1-Second
Slack Reply

@teammate: Can you review my PR?
🎙 "yeah I'll check it right after lunch"
→ "Sure! I'll review it right after lunch. 👍"

Speak your raw thought. HeyDict polishes it to match the tone of the channel, adds the right emoji, and injects it into the active Slack window. Done.

✍️ Tone Matching
Notion Research

Deep Notion
Research

🎙 "Create a task in Notion: review Q1 metrics,
assign to me, due Friday, priority high"
→ Creating task in "Sprint Board"...
✓ Task created · Assigned · Due: Mar 7

Manage entire databases by voice. Create, update, and query your Notion workspace at the speed of thought—no browser tabs, no clicking through menus.

🗂 Database Actions
Open Source & Privacy

Transparency is the
Future of AI.

Your voice is the most personal data you own. HeyDict is open source so you can verify exactly how your data is handled.

Privacy Shield
✓

Fully Open Source Architecture

Built with transparency at its core. Audit the code yourself. No black boxes, no hidden trackers, no trust-me tokens.

✓

Local-First Architecture

HeyDict performs all speech-to-text processing on-device. No audio is ever uploaded to a central server.

✓

Flexible Cloud Fallback

When complex tasks require an LLM, HeyDict sends only the minimal context needed—never your raw audio.

Proactive Intelligence

It Sees Your Screen.
Before You Even Ask.

Context-Aware.
Always Watching.

HeyDict takes silent, memory-only screenshots to understand your active work context. It identifies terminal errors, reads code context, and detects patterns—then suggests fixes or actions proactively.

◆ Detects build errors and suggests fixes
◆ Reads active file context for smarter responses
◆ Screenshots are memory-only—never saved to disk
◆ Works across IDEs, terminals, and browsers
Terminal — zsh
$ cargo build --release
error[E0382]: borrow of moved value: `config`
--> src/main.rs:42:15
|
42 | let app = App::new(config);
| ------ value moved here

⌨️ HeyDict detected a borrow error. Say "fix it" to clone config before the move.
What's Coming

The Roadmap to
Your Personal Jarvis.

HeyDict is just getting started. Here's what's on the horizon for the most ambitious voice agent ever built.

Coming Soon
Memory

Agent Memory

HeyDict learns your phrasing, preferences, and common patterns. It builds a personal model that gets faster and more accurate over time. Your voice, your style—amplified.

In Development
Recipes

Community Recipe Store

Share and discover voice workflows created by the community. One-click install of pre-built action recipes for tools like Figma, Linear, GitHub, and more.

Exploring
Chains

Multi-Agent Chains

Chain multiple actions into complex workflows. "Summarize this PR, post it in Slack, then create a follow-up task in Linear"—all from a single voice command.

Simple Pricing

Free Forever.
Pro When You Need It.

HeyDict's core voice engine is free and open source. Unlock the full agent power stack with Pro.

Free

€0 / forever

The full local voice engine. No limits, no strings.

  • ✓ Unlimited local dictation
  • ✓ On-device AI Engine
  • ✓ Universal text injection
  • ✓ Hybrid semantic routing
  • ✓ Custom hotkeys & modes
  • ✓ Open source & self-hostable
  • — Agent actions (MCP)
  • — Cloud LLM fallback
  • — Proactive screen context
  • — Agent memory & learning
  • — On-device booster
  • — Priority support
Download Free
Most Popular

Pro

€9 / month

Everything in Free, plus the full agent stack.

  • ✓ Everything in Free
  • ✓ Agent actions via MCP
  • ✓ Cloud LLM fallback
  • ✓ Proactive screen context
  • ✓ Agent memory & learning
  • ✓ On-device booster
  • ✓ Multi-agent chains
  • ✓ Community recipe store
  • ✓ Priority support
  • ✓ Early access to new features
Get Pro

Cancel anytime. No lock-in.