OverviewDetailsScreenshotsDevelopmentLinks
Sidekick

Sidekick

Privacy-first local LLM app for macOS — 3.2K+ GitHub stars

2024-11 — Present
SwiftSwiftUIllama.cppCoreMLRAGGraphRAG

A native macOS app that lets users chat with a local LLM that can respond with information from files, folders, and websites — all without installing any other software. Every conversation happens offline and data stays on-device. Sidekick is local-first, with a built-in llama.cpp inference engine for GGUF models like Qwen 3.5, while also accommodating OpenAI-compatible APIs for remote model options.

Key Highlights

  • 3.2K+ GitHub stars, ranked #1 on Google for "mac llm app"
  • Experts & RAG: scoped knowledge domains over local files, folders, and websites with GraphRAG
  • Agentic function calling: sequential tool-call loops (web search, file ops, run_python, email drafting)
  • Deep Research: autonomous multi-step agent that reads 50–80 webpages and synthesizes full reports
  • Persistent memory across conversations for personalized, context-aware responses
  • Canvas: create, edit, and live-preview websites, code, and documents with selective text editing
  • Toolbox: Inline Writing Assistant (Cmd+Ctrl+I in any app), AI Detector, Diagrammer, Slide Studio
  • Image generation via CoreML with automatic prompt classification — no mode switching
  • Advanced rendering: LaTeX math, auto-generated charts from tables, Mermaid diagrams
  • Speculative decoding on Apple Silicon for fast local inference

Screenshots

Sidekick main chat interface

Main conversation interface with local LLM

Development

Project Info

Stack
Swift, SwiftUI, CoreML
Backend
llama.cpp (GGUF)
Platform
macOS (Apple Silicon)
Models
Qwen 3.5, Llama 3, Gemma 4, DeepSeek

GitHub Stars Over Time

Milestones

v0.0.1 — Initial Release2024-10-17

Local LLM chat with file/folder/website indexing and web search

v0.0.10 — Speculative Decoding2024-11-25

Up to 40% faster local inference on Apple Silicon

v0.0.11 — Image Generation2024-12-20

On-device image generation via CoreML (macOS 15.2+)

v0.0.13 — Reasoning Models2025-02-06

Support for DeepSeek-R1 and other reasoning-capable models

v0.0.17 — Diagrammer2025-02-21

Prompt-to-diagram extension with integrated Mermaid preview

v0.0.19 — Slide Studio2025-03-02

Prompt-to-presentation with PDF/PowerPoint export

v0.0.27 — Canvas2025-03-22

Live website/code preview with selective in-place editing

v0.0.32 — Function Calling2025-04-09

Sequential tool-call loops replace the standalone code interpreter

v1.0.0-rc.8 — Memory2025-04-23

Persistent memory across conversations for personalised responses

v1.0.0-rc.12 — Deep Research2025-05-21

Autonomous multi-step agent reading 50–80 pages and synthesising reports

v1.0.0-rc.16 — Extensions Refresh2025-11-17

Reorganised extensions, refreshed UI, expanded model library

v1.0.0-rc.18 — Latest2026-04-02

Latest release candidate; current as of April 2026

Latest Release

1.0.0-rc.181.0.0 Release Candidate 18 — latest release
Last updated: 2026-04-02