PLOS for Mac

Local-first AI workspace for macOS.

English | 한국어 | 日本語

Why PLOS

PLOS combines a native SwiftUI app and a local Python sidecar to run chat, retrieval, and memory workflows primarily on-device.

• Local-first chat and retrieval (RAG) over workspace files
• Policy-gated external web/provider usage
• Multi-layer memory (session/workspace/preferences/pinned)
• Hardware-aware local model catalog
• Conversational direct-first response policy

Runtime Status

• Local inference backends: MLX and llama.cpp (primary path)
• External providers (optional): OpenAI, Anthropic
• Ollama: not integrated as a first-class runtime in current main branch

Architecture

graph TD
  A["SwiftUI macOS App"] --> B["Python Sidecar (FastAPI)"]
  B --> C["Local Inference (MLX / llama.cpp)"]
  B --> D["SQLite (settings, memory, metadata)"]
  B --> E["LanceDB (vector index)"]
  B --> F["Parsers (txt/md/pdf/ocr)"]
  B --> G["External Providers (optional, policy-gated)"]

Loading

Repository Layout

• PLOS/: macOS app (SwiftUI)
• sidecar/local_ai_core/: reasoning, inference, memory, API
• sidecar/tests/: sidecar tests
• PLOSTests/, PLOSUITests/: app tests

System Requirements

• Apple Silicon Mac recommended (M-series)
• macOS 14+
• Xcode 15+
• Python 3.11+
• Optional OCR tools: tesseract, poppler

Quick Start

1) Clone

git clone https://github.com/adgk2349/PLOS-for-Mac.git
cd PLOS-for-Mac

2) Sidecar environment

cd sidecar
python3 -m venv .venv
source .venv/bin/activate
pip install -e .
pip install -e '.[test]'

3) Optional OCR

brew install tesseract poppler

4) Run app

• Open PLOS.xcodeproj in Xcode
• Run PLOS target
• Sidecar lifecycle is managed by the app
• First-run check: if the chat view loads and answers a simple prompt, startup is healthy

Sidecar Standalone (Dev)

cd sidecar
source .venv/bin/activate
export LOCAL_AI_SESSION_TOKEN=dev-token
export LOCAL_AI_DATA_DIR="$(pwd)/data"
uvicorn local_ai_core.main:create_app --factory --host 127.0.0.1 --port 8787

dev-token is an example value for local development only.

Model Tiers (Practical)

• 16GB: 7B/8B class, limited 12B/14B attempts
• 64GB+: 20B/70B class
• 256GB+: GPT-OSS 120B class
• 500GB+: Kimi 2.5 / Qwen 3.5 397B class

Testing

Sidecar

cd sidecar
source .venv/bin/activate
pytest -q

Focused regressions

pytest -q tests/v2/chat/test_v2_pipeline.py tests/test_local_inference_sanitize.py tests/test_memory_service_digest.py

App tests

xcodebuild \
  -project PLOS.xcodeproj \
  -scheme PLOS \
  -destination 'platform=macOS' \
  test

Docs

• CONTRIBUTING.md
• PERFORMANCE.md
• CHANGELOG.en.md
• CHANGELOG.ko.md
• CHANGELOG.ja.md

License

MIT. See LICENSE.