Control Hebrish from your phone with Macro Deck
Macro Deck is a free, open-source stream deck. It runs on your computer, and your phone, a tablet or any browser shows its buttons. With the Hebrish deck, your phone becomes a remote for dictation on Ubuntu or Windows: hold a button to talk, let go and the text is typed where your cursor is. Other buttons start and stop recording with a tap and switch between Sabich, Shakshuka and Falafel.
Macro Deck isn't part of Hebrish; it's made by the Macro Deck project. Daniel uses this deck himself; it was built and tested on 2026-11-16 with Macro Deck 3.0.2 on Ubuntu 26.04.
Why it works well with Hebrish
The Hebrish desktop app has a local API: a small web server on your own computer, at 127.0.0.1:8723, that other programs can use to start and stop dictation and change model. Controllers like Macro Deck are one of the reasons it's there, and physical stream decks and button boxes can use it the same way later.
Each button sends one request from Macro Deck to Hebrish, both on the same computer. Nothing has to have window focus and no keystrokes are faked, so it works on Wayland, where key-sending tools often don't. Your phone only sends the button press to Macro Deck; the audio is recorded by the computer's microphone and never leaves it.
The deck
Two columns and four rows, so the buttons are big on a phone held upright, with the most-used one at the bottom where your thumb is.
| Button | What it does |
|---|---|
| Hold to talk | Push to talk: records while you hold it; letting go stops and types the text. It uses Macro Deck's Touch start and Touch end triggers. |
| Tap: start / stop | Tap to start recording, tap again to stop and type. |
| Stop | Stops recording and types. A safety net if a hold was interrupted. |
| Sabich | Switches to Sabich, the everyday model. |
| Shakshuka | Switches to Shakshuka, the streaming model: text appears while you speak. |
| Falafel | Switches to Falafel, the model for short voice commands. |
A model button only switches to a model you've already downloaded on Hebrish's Models tab; it never starts a download.
Before you start
- Hebrish desktop 1.018 or later, on Ubuntu or Windows. On Ubuntu:
sudo apt update && sudo apt install hebrish. - Macro Deck 3 on the same computer, from macro-deck.app (on Ubuntu and Debian, its APT repository). Macro Deck 2 and its old Android app don't work with this deck.
- The local API turned on. In Hebrish, open Settings → Local API and tick both Serve transcription to apps on this computer and Let agents read history and control the app. The second switch is what lets the deck start dictation; it also lets programs on your computer read your transcript history, so turn it on only if you're happy with that.
Install the deck
Three ways, from quickest to most hands-on.
1. Import the template
- Download hebrish-dictation.macroDeckProfile.
- In Macro Deck, open the profile menu, choose Import profile and pick the file. It arrives as the profile Hebrish Dictation, icons included.
- On your phone, open Macro Deck's deck (scan the pairing code Macro Deck shows) and switch to that profile.
The same deck as a readable spec, for building your own or adapting it: hebrish-dictation.deck.json.
2. Let an agent set it up
The set-up-hebrish-macro-deck agent skill turns on the local API, checks which models you have, and installs the deck, either by import or by building it through Macro Deck's built-in MCP server (create a token in Macro Deck under Settings → AI clients and connect your agent first). Paste this into Claude Code, Codex, Gemini CLI or another coding agent:
Set up a Macro Deck for Hebrish on this computer using https://raw.githubusercontent.com/Hebrish-ASR/Hebrish-Agent-Skills/main/skills/set-up-hebrish-macro-deck/SKILL.md
The agent won't press the dictation buttons for you: they'd record your microphone and type into whatever window is open. It tests the connection by re-selecting the model you already have.
3. By hand
In Macro Deck, make a profile and add six buttons. Each one gets an HTTP → Send HTTP Request action, method POST, body type JSON:
| Button | Trigger | URL | JSON body |
|---|---|---|---|
| Hold to talk | Touch start | http://127.0.0.1:8723/v1/app/dictation | {"action": "start"} |
| Touch end | http://127.0.0.1:8723/v1/app/dictation | {"action": "stop"} | |
| Tap: start / stop | Short press | http://127.0.0.1:8723/v1/app/dictation | {"action": "toggle"} |
| Stop | Short press | http://127.0.0.1:8723/v1/app/dictation | {"action": "stop"} |
| Sabich | Short press | http://127.0.0.1:8723/v1/app/models/select | {"id": "sabich-rolling", "download": false} |
| Shakshuka | Short press | http://127.0.0.1:8723/v1/app/models/select | {"id": "shakshuka", "download": false} |
| Falafel | Short press | http://127.0.0.1:8723/v1/app/models/select | {"id": "falafel", "download": false} |
If you changed the port in Hebrish's Settings → Local API, use yours instead of 8723.
Check it works
Click into a text field on the computer, hold Hold to talk on your phone, say a sentence and let go. If nothing is typed, run this on the computer while you hold the button:
curl -s http://127.0.0.1:8723/v1/app/status
It should say "state": "recording". A connection error means the local API is off or Hebrish isn't running; a 403 about control means the second switch is off. Anything else: open an issue.