Local AI Chat (Cat)

AI that stays at home. Run models locally on Apple Devices & clean your menu bar by having all productivity tools in one cat.

Comment

Inspiration

Local AI Cat (Chat) was initially called Seek Deep but apple didn't like that, it was my first foray into local AI and MLX back in the time pre coding agents when it was still pasting in between chat and vs code.

After getting the MVP out, very rudimentary, exploring whether I could try and get deepseek to pretend to think like a cat via prompting only (it didn't) I neglected this app for a bit due to other priorities, but over the last year I've been building on it intensely.

The idea is basically instead of getting another mac menu bar app to hide all my menu bar apps, what if I just make one super app that brings it all together, with any tweaks that I personally want to add on top of nice MIT OSS projects.

What it does

Many things...

Local LLM Chat + Server Transcription Engine - multiple models, on hotkey press, in background, ad hoc Translation Engine - multiple models, conversational, literal, lyrics VNC + SSH + 3D Cat Terminal Eye Care Reminder + Fitness Pal Keyboard Cleaning Mode Backlight Brightness Management External Monitor & Brightness Management

How we built it

We built the Cat natively with SwiftUI, MLX, WhisperKit, FluidAudio, and Apple frameworks. Recently with Claude building specifications and coordinating work, with Codex agents implementing, testing, and reviewing.

I've been extensively refactoring the entire codebase focusing on modularity and testability using Codex actively to deslop as much as possible.

Oh also around 20-30k of tokens (if at API billing rates)

Challenges we ran into

Developing for Apple devices is hard and the dx is very often painful. Good testing loops are hard to come by and need many devices at my disposal.

Accomplishments that we're proud of

Very happy with the entire app, even though stability throughout all features and full completion of them all is still pending, I've been refactoring aggressively and making my future work cleaner by the day, learning extensively along the way.

What we learned

Claude as orchestrator and Codex as executor work very nicely together. /goal is your friend, and my fork of CMUX that has enhanced terminal conducting allows for observability of subagents nicely and direct intervention when I desire.

What's next for Local AI Chat (Cat)

Finish all the features, especially the Cat Broadcast Studio (CatBS) so that I can begin streaming my development to help people learn.

Built With

Share this project:

Updates