windows · mac · linux
flicky
an ai buddy that lives on your desktop.
100% free · open source · bring your own keys
the models got really good, and we're all still typing at them in a chat box in a browser tab. that felt backwards.
so we put one on your screen instead. it looks at what you're doing, talks back out loud, and points at the thing it's talking about — no pasting screenshots, no describing where the button is.
flicky is a from-scratch, cross-platform take on farza's clicky, built by julio so the rest of us get one too.
how it works
01 hear you.
hold to talk, or tap to toggle. groq whisper turns what you said into text before you finish letting go.
02 see your screen.
a screenshot goes with every question — claude, gpt, or a local model reads it, so you never describe anything twice.
you: what am i looking at?
flicky:that's the vercel deploy log — the red line is a missing env var.
03 speak back.
elevenlabs voice, or text-only if you would rather read. the stream window mirrors every word.
04 point at things.
multi-step answers become numbered walkthroughs, and the blue cursor flies to the exact pixel each time.
what you get
local by default
chats and keys are encrypted on your own machine. nothing lives on our servers, because there aren’t any.
your choice of brain
claude sonnet or opus 4.6, the gpt-5 family, or any local / openai-compatible endpoint — ollama, lm studio, whatever you run.
never runs out of context
long conversations auto-compact into a summary, so a single chat can just keep going all day.
types for you
opt in and flicky types straight into the focused field — otherwise the answer lands on your clipboard.
stream window
a floating transparent panel mirrors the live q&a. scroll it, select it, copy straight out of it.
guided setup
a three-minute wizard tests each key, your shortcut and your mic before it lets you finish.
get flicky.
100% free. every tagged release is built and published on github.
you'll need keys for anthropic or openai (or none, with a local model), groq for speech-to-text, and optionally elevenlabs for a voice. add them in the app — they never leave your machine.
questions
is it private?
yes. everything runs locally — your chats stay on your machine and your api keys are encrypted at rest. there is no flicky backend for anything to be sent to. the only network calls are the ones you configure, straight to the providers you picked.
what does it cost?
nothing. flicky is free and mit licensed. you pay your model and voice providers directly for what you use — and if you run a local model through ollama or lm studio, that part costs nothing at all.
does it work on windows?
yes — that's the whole point of this project. hold-to-talk, the tray behaviour, the mic permission flow and the setup wizard were all built and tested on windows, not bolted on afterwards. linux works too.
how is this different from clicky?
clicky is farza's original macos app and the whole inspiration — he invented this interaction. flicky is an independent electron rebuild so windows and linux folks can have it too. if you're on a mac, go use clicky.
do i need an account?
no — no sign-up, no server, no telemetry by default. download it, add your own keys, done.
credit where it's due
Flicky is an independent, cross-platform reimagining of Clicky by Farza — the original macOS app that invented the hold-a-hotkey, get-a-pointing-cursor interaction. Every bit of that vibe is his. Flicky rebuilds the same idea in Electron so people on Windows and Linux can try it too.
If you liked Flicky, also go star farzaa/clicky.