Skip to content

Ravenkao/aya-local-copilot

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 
 
 

Repository files navigation

Aya — Local AI Copilot for macOS

Turn any text in macOS into AI-enhanced output — instantly, locally, and without leaving your workflow.

Aya is a system-level AI assistant that enables users to rewrite, translate, and summarize text across any application using a global hotkey — powered by a local multilingual model.

Learn more about Tiny Aya: https://cohere.com/blog/cohere-labs-tiny-aya


🎥 Demo

https://drive.google.com/file/d/1JuajpfD5qvZozoa-_jAfl_jq95zJB6AO/view?usp=sharing


🧠 Why I Built This

Most AI writing tools today (e.g. Copilot, Grammarly) require:

  • switching context between apps
  • repetitive copy/paste workflows
  • sending sensitive data to cloud APIs

Aya explores a different interaction model:

AI as a system-layer capability — not just another app.

This is especially important in privacy-sensitive environments (e.g. B2B, enterprise workflows) where sending internal content to external APIs is not acceptable.


🌍 Multilingual by Design (Tiny Aya)

Aya leverages Tiny Aya, a multilingual model developed by Cohere Labs.

Tiny Aya supports 70+ languages, including many low-resource languages, enabling:

  • Cross-language communication
  • Multilingual rewriting
  • Translation without external APIs

This makes Aya particularly valuable for:

  • global teams
  • international operations
  • multilingual product environments

🔒 Why Local (B2B Perspective)

Aya is built as a local-first AI system, which directly addresses enterprise concerns:

Privacy & Compliance

  • No data leaves the device
  • No third-party API calls
  • Suitable for internal documents, Slack messages, Notion content

Cost Control

  • No per-token API cost
  • Predictable infrastructure usage

Reliability

  • Works offline
  • No dependency on external services

This design aligns with how many B2B teams evaluate AI adoption:
control, compliance, and predictability over convenience.


🚀 Features

  • Fix grammar instantly
  • Translate (multilingual, 70+ languages)
  • Rewrite professionally
  • Summarize text
  • Global hotkey (Cmd + Shift + .)
  • Works across any app (Notion, Slack, browser, etc.)
  • Fully local inference (Tiny Aya via aya-cli)

⚙️ How It Works

  1. User selects text in any application
  2. Aya captures the selection via clipboard
  3. Sends request to local AI model (aya-cli)
  4. Receives structured output
  5. Replaces or returns improved text

🏗️ Architecture

User Selection ↓ Clipboard Capture ↓ macOS Menu Bar App (Swift) ↓ Local API (aya-cli) ↓ Tiny Aya (llama.cpp backend)


🛠️ Tech Stack

  • Swift (macOS menu bar app)
  • aya-cli (local LLM runtime)
  • llama.cpp (model inference engine)
  • Tiny Aya (multilingual model)

📦 Setup

1. Install aya-cli

https://github.com/Cohere-Labs/aya-cli

2. Run local server

aya-cli serve

3. Run macOS app

see demo

📌 Current Limitations

Clipboard-based interaction (paste simulation) No UI for mode switching yet Requires local model setup

🚧 Future Work

Direct inline replacement (no paste simulation) Mode selection UI (rewrite / translate / summarize) Prompt customization Memory / personalization layer Performance optimization for real-time usage Packaging as installable macOS app

🙏 Acknowledgements

This project uses Tiny Aya via aya-cli by Cohere Labs: https://cohere.com/blog/cohere-labs-tiny-aya https://github.com/Cohere-Labs/aya-cli

About

A local-first macOS AI copilot that rewrites, translates, and summarizes text across any app using Tiny Aya—fully on-device with no cloud dependency.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages