A Windows fork of tajchert/tink-agent.
Turn a Teenage Engineering EP-2350 "Ting" mic into a voice-and-button controller for AI coding agents.
Hold the handle, talk to your agent, and let go. Press the mic buttons to approve, reject, interrupt, move through modes, or send little command macros.
This fork keeps the same audio/DSP pipeline (tone detection + push-to-talk STT) and ships a Windows system-tray app with Settings, onboarding, and Map-a-Button UI styled after the original macOS app.
Built for fun in Cursor over a few hours after stumbling across the original repo. Yes, i know..JAVCM (just another vibe coded monstrosity)
As I leveraged the prebuilt Ting config and wav's from the original project, they won't be included within this repository. Please visit tajchert/tink-agent to collect the files needed to setup the Ting.
I welcome someone taking this forward, consider it inspiration for a packaged windows variant ;)
- Windows 10/11
- Python 3.10+ (3.12+ recommended)
- Teenage Engineering EP-2350 Ting — battery-powered; line-out into a USB audio adapter
- Speech-to-text, one of:
- EasyWhisperUI (recommended / default) —
portable/installer wraps whisper.cpp with GPU (Vulkan). Run it once so it builds
whisper-cliunder%APPDATA%\easy-whisper-electron\whisper-workspace, download a model (e.g.base.en), then in Tink Agent Settings choose EasyWhisperUI. Browse to the portable folder orwhisper-workspaceif auto-detect misses it. whisper-clion PATH (whisper-cppbackend)whisperfrom openai-whisper- a custom command that accepts a WAV and prints/writes text
- EasyWhisperUI (recommended / default) —
portable/installer wraps whisper.cpp with GPU (Vulkan). Run it once so it builds
The Ting does not send button events over USB. You copy tone samples from
ting-config onto TINGDISK; tink-agent then detects those tones
in the analog audio stream with a Goertzel detector and turns them into keystrokes.
See TING.md for device setup. First-run onboarding can also Copy the
files for you when the volume is mounted.
One-time USB-C setup:
- Connect Ting over USB-C so
TINGDISKmounts (Windows volume label TINGDISK). - Copy the contents of ting-config to the root of that drive.
- Restart the Ting. Disconnect USB-C for normal use; use batteries + line-out only.
Easiest: double-click / run .\run.bat (creates .venv, installs deps, starts the app).
Or install once, then run:
py -3 -m venv .venv
.\.venv\Scripts\pip install -r requirements.txt
.\.venv\Scripts\python -m tink_agentA tray icon appears. Open Settings… from the tray menu.
Installer script (same venv + deps; optional Start at login):
.\packaging\install.ps1
.\packaging\install.ps1 -StartAtLoginRemove the Startup entry (does not delete the repo or config):
.\packaging\uninstall.ps1Paths:
Config: %USERPROFILE%\.tink-agent\config.json
Run log: %LOCALAPPDATA%\TinkAgent\TinkAgent.log (Startup launches)
Activity log: %LOCALAPPDATA%\TinkAgent\TinkAgent-activity.log (opt-in)
- Allow microphone access when Windows prompts (first capture).
- Keystroke injection via
pynputmay be blocked by some antivirus products — allowlist the Python / tink-agent process if actions do nothing.
Useful fields in config.json:
device_name— input device name substring (defaultUSB PnP Sound Device; also tryUSB Audio Device)stt_backend—easy-whisper-ui(default),whisper-cpp,openai-whisper, orcustomstt_model— whisper model path/name (ggml/gguf)easy_whisper_dir— EasyWhisperUI portable folder orwhisper-workspace(optional; auto-detects%APPDATA%\easy-whisper-electron\…)whisper_cpp_binary— absolute path towhisper-cli.exe(optional)stt_language— whisper language code (defaulten)slot_actions— maps tone slots 1–8 to action IDstarget_apps— optional list of exe/path/name substrings; empty = any focused applog_activity— opt-in TSV activity log (TinkAgent-activity.logunder%LOCALAPPDATA%\TinkAgent\)
Most everyday settings are in the Settings window.
enter, escape, esc_esc, ctrl_c, ctrl_d,
tab, shift_tab, shift_enter,
up, down, left, right,
space, backspace, ctrl_v, cmd_v,
key_1, key_2, key_3,
type:yes, type:continue, type:/clear, type:/compact,
noop
cmd_v is an alias for Ctrl+V paste on Windows.
AudioCapture
-> Engine.handle_block()
-> ToneDetector.process()
-> ActionRouter.fire_slot()
-> VoiceGate.process()
-> Transcriber.transcribe()
-> ActionRouter.type_text()
| File | Role |
|---|---|
tink_agent/__main__.py |
Entry: python -m tink_agent → qt.main() |
tink_agent/qt/app.py |
Tray lifecycle (TinkQtApp), shared setters |
tink_agent/qt/settings.py |
Settings window |
tink_agent/qt/onboarding.py |
First-run “Set up TINK” |
tink_agent/qt/button_map.py |
Map a Button (8 slots ↔ actions) |
tink_agent/qt/theme.py |
Cream/orange/green Qt stylesheet + Archivo |
tink_agent/audio.py |
PortAudio capture and device scanning |
tink_agent/detector.py |
Goertzel tone detection and energy VAD |
tink_agent/engine.py |
Pipeline orchestration (Win32 frontmost app) |
tink_agent/transcribe.py |
External STT command wrapper |
tink_agent/easy_whisper.py |
EasyWhisperUI path/model discovery |
tink_agent/actions.py |
Slot-to-action routing and keyboard output |
tink_agent/setup_checks.py |
Onboarding checks (TINGDISK volume label) |
tink_agent/startup.py |
Windows Startup folder helpers |
tink_agent/activity_log.py |
Opt-in TSV activity log |
.\.venv\Scripts\pytest -qThis is an independent Windows fork of the MIT-licensed tajchert/tink-agent macOS app. It is not affiliated with Teenage Engineering.
MIT. See LICENSE.