sttui 0.3.0a4

Linux-first speech-to-text terminal UI

sttui: Speech To Text in your terminal

No browser. No Web UI. Fast speech-to-text with the best models

https://github.com/user-attachments/assets/252ba77e-d3f3-4689-bcc1-77f536f10c60

Setup

pip install sttui

(or if you have uv installed, uvx sttui)

Then, you must have an account on openrouter, and get an API key.

To register it, run:

sttui auth

Storage of your key

Your key will be stored inside ~/.config/sttui/auth.json

Make sure you don't commit this file !

Config

When you first start the app, a config file is created at: ~/.config/sttui/config.toml

You can specify the default model (without the openrouter prefix), the prompt, and the maximum audio length.

[transcription]
model = "mistralai/voxtral-small-24b-2507"
prompt = """
You are a helpful assistant that can hear audio and write text.
Return a transcription of the user audio as json. If the user request is empty, return null.
<format>
{
  "transcription": ""
}
</format>
<format>
{
  "transcription": null
}
</format>
"""
max_seconds = 600

⚠️ Make sure that the prompt asks the model to answer in this json format, it's the one expected by sttui

Commands

# Start interactive dictation TUI
sttui

# Equivalent explicit run command
sttui run

# Show CLI help
sttui --help

# Set or update API key
sttui auth

# TUI + write transcript to stdout on Enter
sttui --stdout

# Override model and recording cap for this run
sttui --model google/gemini-2.5-flash --max-seconds 120

# Use a custom config file
sttui --config ~/.config/sttui/config.toml

# Record, transcribe, and send to an HTTP endpoint
sttui --send-post https://example.com --send-body '{"text": $0}'

# Send transcript to a shell command
sttui --send-command 'xargs -I {} notify-send "{}"'

# Send transcript to a Unix socket (e.g., pi coding agent)
sttui --send-socket /run/user/1000/pi/sttui.sock --send-body '{"message": $0}'

# Chain multiple sends (with 1s delay between them)
sttui --send-post https://example.com/foo --send-body '{"a": $1}' \
           --send-post https://example.com/bar --send-body '{}' \
           --send-delay 1000

# Background lifecycle (no TUI)
sttui background start
sttui background stop
sttui background toggle

# Same with desktop notifications
sttui background --notify start

Send Command Templates

In --body templates, use $0 for the full transcript, $1/$2/etc. for individual parts.

Values are JSON-escaped automatically when a --body template is provided.

All recordings and transcripts are stored in ~/.local/share/sttui/recordings/.

Integrations

AI coding agents (opencode, pi)

Dictate directly into coding agents.

pi coding agent

Dictate directly into pi using sttui's socket integration.

Copy the extension to your pi extensions folder:

mkdir -p ~/.pi/agent/extensions
cp integrations/pi.ts ~/.pi/agent/extensions/sttui.ts

After starting pi, you'll see the sttui command in the chat.

Contributing

This is a side-project of mine. I must admit there is mostly AI-generated code, but I try to review and ensure good practices.

I don't have strong opinions about how this project should evolve. If you find it useful, feel free to contribute !