Developer preview

Install path (early preview)

Speech Layer is the voice front door on your desktop. You speak, your machine turns that into text, work can fan out to coding agents, and a local voice can answer back. Everything that matters stays on your computer unless you turn on a cloud option.

Release expected August 2026. This page is for people already poking at the stack. General availability is not open yet — join the waitlist for the launch notice. If a step already looks done on your machine, skip it.

The stack in plain words

What you are installing

Think of four jobs that work as a chain. You do not need every piece on day one, but each one has a job:

Speech Layer (this product)
Listens for your wake phrase, runs the desktop voice loop, and hands text to tools. After install you use the speechlayer command in a terminal.
Whisper-class speech-to-text (hearing)
Turns your spoken words into text on this machine. The marketing stack points at OpenAI Whisper. A production installer path has also built whisper.cpp (a fast local build of the same idea). Either way: text from audio, locally.
herdr (the workbench; STT often hears this as “herda”, “heard”, or “herd”)
A terminal workspace manager. Named panes hold separate agent sessions so several jobs can run side by side. Official install: herdr.dev install docs.
pi (the coding agent in each pane)
The agent harness that actually does coding work inside a herdr pane. Official install: pi.dev.

Optional later: a cloud model key (for example OpenRouter) if you do not run a fully offline model path, plus text-to-speech so the machine can speak answers. Voice dictation alone can work before every agent piece is wired.

Prep

Before you start

Use Linux or macOS
Those are the desktops Speech Layer targets. Windows is not a first-class path on this site.
Open a terminal
Every command below is typed in a terminal window. On macOS that is Terminal or iTerm. On Linux it is your normal terminal app.
Have a microphone that the OS already sees
If other apps cannot record you, fix that in system settings first. Speech Layer cannot hear a mic the OS does not allow.
Know that peers may install separately
herdr and pi publish their own installers. Whisper has its own too. A full Speech Layer installer is meant to bring the voice core up; if a peer is still missing, use the matching section below.

One command

Quick path: run the installer

The homepage one-liner is the intended fast path on both Linux and macOS:

# Linux and macOS — intended installer entry
$ curl -fsSL https://speechlayer.com/install.sh | sh

Honest status of that URL

At the time this guide was written, https://speechlayer.com/install.sh was not returning an installer file (the site answered with a normal web page instead). There is also no install.sh checked into this marketing tree. If the pipe fails or you only get HTML, stop: do not force the command. Use the checklists below for the pieces you can install from their own sites, or contact us and say install.sh was missing.

When a real installer is live, the product README expects roughly:

Speech Layer home under ~/speechlayer/
Config under ~/.config/speechlayer/
Commands such as speechlayer on your PATH (often via ~/.local/bin)
A chance to start the background listener (the “daemon”) and a small on-screen indicator

Changelog notes from earlier portability work also describe the installer building local transcription (including a whisper.cpp path on Linux with build tools present). Treat that as “installer may build STT,” not as a guarantee on every OS until your own check passes.

Verify

Check that it worked

Run these one at a time. A green path is “command found and it prints help or status.” A missing command is normal if that piece never installed — jump to the matching section.

1. Speech Layer CLI

$ command -v speechlayer
$ speechlayer status
$ speechlayer --help

You want command -v to print a path, and status / --help to speak Speech Layer’s own words. If the daemon is “not running,” that is a start problem, not always an install failure — see First run.

2. Transcription tools (Whisper family)

# any one of these may be enough, depending on how STT was packaged
$ command -v whisper
$ command -v whisper-cpp
$ command -v whisper-cli
$ ls ~/speechlayer 2>/dev/null
$ ls ~/.cache/speechlayer 2>/dev/null

Exact binary names vary by package and version. What matters: something on the box can turn a short recording into text, and Speech Layer’s own speechlayer test (when the daemon is up) is the product-level smoke check.

3. herdr

$ command -v herdr
$ herdr --help

4. pi

$ command -v pi
$ pi --help

Skip joy for a piece that already passed. Only fix the lines that print nothing or “command not found.”

If the installer did not bring hearing

If Whisper-class speech-to-text is missing

You can skip this whole section if speechlayer test already transcribes, or if a whisper binary clearly exists.

Prefer the Speech Layer installer once it is live
Changelog burn-in notes describe install.sh building whisper.cpp with normal compiler tools on Debian/Ubuntu-class systems. If install.sh starts working for you, re-run it before hand-building STT.
Manual path: official Python Whisper (peer product)
From the OpenAI Whisper README, the published install line is:
```
$ pip install -U openai-whisper
```
That stack also needs ffmpeg on the system. Examples from the same README:
```
# Debian / Ubuntu
$ sudo apt update && sudo apt install ffmpeg

# macOS (Homebrew)
$ brew install ffmpeg
```
Do not copy a whisper binary from another distro
Portability notes on this project warn against transplanting cross-distro whisper builds. Build or install on the machine you will actually use.
Re-check
```
$ command -v whisper
$ speechlayer test
```

If the workbench did not land

If herdr is missing

Skip if command -v herdr already prints a path.

herdr is a separate product. Speech Layer drives it; it does not replace it. Install from herdr’s own docs:

Stable curl installer (from herdr.dev)

$ curl -fsSL https://herdr.dev/install.sh | sh

Or Homebrew (also offered on herdr’s site)
```
$ brew install herdr
```
Confirm
```
$ command -v herdr
$ herdr --help
```

Voice control talks to a running herdr session. After the binary installs, open herdr the way you usually start a workspace, then try Speech Layer again.

If the agent harness did not land

If pi is missing

Skip if command -v pi already works.

pi is also a separate product. Official floating installers from pi.dev include:

curl installer

$ curl -fsSL https://pi.dev/install.sh | sh

Or npm (same site; ignore-scripts flag as published there)

$ npm install -g --ignore-scripts @earendil-works/pi-coding-agent

Confirm
```
$ command -v pi
$ pi --help
```

You still need a way for pi to reach a model (local llama.cpp style setup, or an API key such as OpenRouter). That is model config, not the install of the pi binary itself. Key and provider setup belong in pi’s own docs once the command exists.

Smoke test

First run

Only after the checks you care about look sane:

Start Speech Layer
```
$ speechlayer restart
$ speechlayer status
```
Product docs also show speechlayer start on the homepage snippet; the shipping CLI help on a rebranded tree lists restart as the full start path. If one verb is unknown, use the other from speechlayer --help.
End-to-end self test
```
$ speechlayer test
```
Wake word (default on the rebranded product tree)
Say “speech layer” or “hey speech layer.” The documented default action opens a terminal path and types what you said (dictation into the terminal). Give the mic a second; do not shout.
Optional: ask the machine to speak
```
$ speechlayer-say "Speech Layer is ready"
```
That is the product TTS wrapper name on the current tree. A missing voice engine is a TTS follow-up, not a reason to reinstall herdr.
Optional full loop
With herdr open and pi available, speak a simple agent task. Speech Layer routes; herdr holds panes; pi works; a voice can read back. If only dictation works, you still have a useful speech layer — agents are the next rung.

Help

Still stuck

Collect three lines of truth

$ uname -a
$ command -v speechlayer herdr pi whisper
$ speechlayer status

Write us
Use the contact form. Paste those three outputs and say which step failed (installer URL, missing binary, wake word silent, agents not opening).

Join the waitlist Contact How the stack fits together