This is a native Windows port of the publicly available, MIT-licensed Clicky
snapshot in ../upstream-clicky. It is a tray-first
desktop companion that follows the original public interaction:
- Open the Clicky panel from the tray.
- Complete the local permission/setup flow.
- Hold Ctrl + Alt to talk; release to submit.
- Clicky records microphone audio while held, captures each display once only
after release, sends the transcript and screenshots to your Worker, speaks
the answer, and can point to a returned
[POINT:…]target.
Use the self-contained Windows build:
release\Clicky.Windows.exe
The release directory contains the executable and its bundled .NET runtime;
copy the whole directory together. It does not require a machine-wide .NET
runtime installation.
The app intentionally does not accept or embed provider keys. Set an owner-operated proxy URL before launching:
$env:CLICKY_WORKER_URL = "https://your-worker.example/"
Start-Process .\release\Clicky.Windows.exeThat Worker must implement the public upstream routes:
POST /transcribe-tokenfor an AssemblyAI streaming tokenPOST /chatas an Anthropic Messages streaming proxyPOST /ttsas an ElevenLabs MP3 proxy
The MIT upstream Worker is available at
../upstream-clicky/worker. Deploy a copy you
control with your own provider credentials. The desktop app never sends raw
provider keys to a service and does not reuse the upstream's analytics or
email endpoints.
- No desktop capture runs while idle.
- Capture happens only after Ctrl+Alt is released and only for the current interaction.
- Clicky's windows are marked to exclude themselves from normal Windows screen capture. The app also hides its overlay for its own capture pass as a fallback.
- Windows secure desktop, UAC prompts, DRM-protected content, and elevated applications can remain unavailable to ordinary desktop capture.
The upstream repository is an older open-source Clicky snapshot. Its README states that newer HeyClicky work is private, so this port faithfully targets the public companion: tray UI, setup panel, Ctrl+Alt voice interaction, multi-display capture, Worker-backed Claude/AssemblyAI/ElevenLabs integration, and cursor pointing. It does not claim the private background-agent features advertised on the current heyclicky site.
See ATTRIBUTION.md and UPSTREAM_LICENSE.md
for the upstream attribution and license.