Clicky
Clicky is Farza's open-source macOS AI teacher/assistant: a menu-bar companion that can see the screen, listen through push-to-talk, speak through TTS, and point at UI coordinates. Source: GitHub
farzaa/clicky, 2026-06-30
What Matters
Clicky is useful as product/architecture evidence for Agent Product Surface, not as a default installed tool. The durable pattern is ambient agent UX: the agent lives next to the cursor, sees enough context to help, speaks back, and can point at a precise on-screen target. That makes it a stronger reference than a generic chat overlay when designing screen-aware tutors, coding companions, or local desktop copilots. Source: GitHub farzaa/clicky, 2026-06-30
The newer workflow demo shows the private/live Clicky direction as a desktop companion rather than only the open-source repo. The video moves from Farza talking to a Mac desktop with a compact dark control panel: Home/Agents navigation, hold-to-control and question-to-talk affordances, cursor controls, app integrations, and an "Unlock Clicky" action. Treat this as product-shape evidence for voice-first agent control surfaces; the public farzaa/clicky repo remains the versioned architecture reference. Source: X/@FarzaTV, 2026-05-16; Source: local video contact sheet, reviewed 2026-07-02
The recovered HeyClicky vision doc makes the product thesis explicit: Clicky is a consumer-facing Mac interface for frontier models, intended to move AI work out of chat boxes and terminal-only workflows. Farza reported the five-week-old product had 250,000 messages sent, 50,000 agents spawned, 3,500 weekly active users, and 185 paying customers. Keep those as founder-reported traction, and keep the open-source repo as architecture authority because the live product has no public package or release channel. Source: X/@FarzaTV, 2026-05-22; Source: heyclicky vision Google Doc, recovered 2026-07-04
The HeyClicky Figma Motion demo adds a concrete screen-control case: the user asks the agent to teach Figma Motion and then move a square, while the video shows Figma's Motion controls, red path/target overlays, and canvas movement. This is stronger product evidence than a generic "AI buddy" claim because the agent is operating inside a professional design tool with visible intermediate state. Source: X/@heyclicky, 2026-06-26; Source: local video contact sheet, reviewed 2026-07-04
Routing Summary
Use Clicky as screen-aware, voice-first desktop-agent UX prior art. Do not treat it as a default Kevin runtime or reusable skill. Route actual desktop automation through the active browser/computer-use stack unless Kevin is explicitly building a macOS screen-aware companion. Source: X bookmark artifact review, 2026-06-30
Architecture
The repo is a native macOS app plus a Cloudflare Worker proxy:
| Layer | Implementation |
|---|---|
| Desktop shell | Swift menu-bar app with NSPanel surfaces and global shortcut behavior. |
| Context capture | ScreenCaptureKit screenshot/screen-content access plus microphone permissions. |
| Speech input | AssemblyAI real-time transcription through a proxy token endpoint. |
| Reasoning | Claude streaming SSE response path. |
| Speech output | ElevenLabs TTS playback. |
| Pointer guidance | Claude emits coordinate tags such as [POINT:x,y:label:screenN]; the app animates a cursor overlay to that location. |
| Secret boundary | Cloudflare Worker holds Anthropic, AssemblyAI, and ElevenLabs keys; the app calls the Worker instead of embedding provider keys. |
Source: GitHub farzaa/clicky README, 2026-06-30
Agent Setup Pattern
The repo explicitly asks users to start Claude Code, clone the repo, read CLAUDE.md, and have the agent walk through Worker secrets, proxy URLs, and Xcode setup. That is the reusable lesson: serious open-source apps can ship an agent-readable setup path as first-class onboarding. Kevin's equivalent is AGENTS.md, SKILL.md, doctor checks, and source-owned setup scripts. Source: GitHub farzaa/clicky README, 2026-06-30
Version Snapshot
Current source authority on 2026-06-30: GitHub farzaa/clicky, MIT license, default branch main, HEAD a80fa80721a8aebe51a170a7780705024ebc6e46, tags v1 and v2, no GitHub release, 6,946 stars, 1,342 forks, last pushed 2026-04-28. The README says new private Clicky work moved to heyclicky.com, so treat this repo as a public reference implementation rather than the live product source. Source: GitHub API and git ls-remote, 2026-06-30
Routing
Use Clicky when designing:
- screen-aware agent surfaces;
- voice-first local assistants;
- coordinate-pointing UX;
- Cloudflare Worker provider-key proxy patterns;
- agent-readable open-source setup docs.
Do not create a local executable skill from Clicky yet. The repo is an app reference, not a repeated Kevin workflow. If Kevin starts building a macOS screen-aware companion, create a project-specific skill from the setup and architecture notes then. Source: X bookmark artifact review, 2026-06-30
Timeline
- 2026-04-08 | Farza open-sourced Clicky and encouraged people to build from it. Source: X/@FarzaTV, 2026-04-08
- 2026-04-27 | README update said the public codebase remains open source, while newer Clicky work moved private. Source: GitHub
farzaa/clickyREADME, 2026-06-30 - 2026-06-30 | Focused tool page created during one-by-one X bookmark artifact review; split Farza's repo from Kevin's separate Clicky IRL concept. Source: raw/x-bookmarks/enriched/2041691382008705321.json
- 2026-07-02 | Added Farza's voice-first workflow demo as product-shape evidence: a compact desktop control panel with Home/Agents navigation, voice/control affordances, cursor controls, app integrations, and an unlock path. No new version snapshot applies because the live/private demo has no public package or release channel. Source: X/@FarzaTV, 2026-05-16; Source: local video contact sheet, 2026-07-02
- 2026-07-04 | Recovered the HeyClicky vision doc linked from the May 22 recruiting post and added the consumer-interface thesis plus self-reported early traction. No new version snapshot applies because the live HeyClicky product has no public package or release channel. Source: X/@FarzaTV, 2026-05-22; Source: heyclicky vision Google Doc, recovered 2026-07-04
- 2026-07-04 | Added the HeyClicky voice-to-Figma-motion video as design-tool screen-control evidence: voice prompts move a square on a Figma Motion canvas with visible path/target controls. No new public version snapshot applies. Source: X/@heyclicky, 2026-06-26; Source: local video contact sheet, 2026-07-04