Fol

🤖 FOL — My Personal AI Assistant for macOSI’m building FOL, a personal AI assistant for macOS inspired by the idea of a real JARVIS.FOL is designed to live directly on my MacBook through a native Notch UI and interact with my computer, applications, browser, files, memory, and tools.🚀 What I’m buildingFOL can already:🖥️ Control the desktop and interact with applications🌐 Control Chrome and Safari👁️ Understand what is happening on the screen🎙️ Support voice interaction🧠 Store and retrieve long-term memory🗂️ Integrate with Obsidian as a personal knowledge base🤖 Use an agent loop to select and execute tools✉️ Work with productivity tools such as Gmail and Google Calendar⚡ Stream responses through SSE🔐 Require confirmation before risky actions🧩 Use a centralized Tool Registry instead of scattered tool definitions🔄 Automatically fall back between LLM providers when one fails🏠 Support local models through Ollama🏗️ ArchitectureThe project is built around several services:SwiftUI → Orchestrator → LLM Router → Tool Registry → Agent Server / Memory / Productivity / FOL CoreThe current architecture uses:Python + FastAPILiteLLMSwiftUIReact / Next.jsPyAutoGUIChrome CDPOllamaObsidianSSE / WebSocket🧠 Recent progressI’ve been rebuilding the architecture phase by phase.Phase 3: unified LLM routing and deterministic fallback.Phase 4: centralized ToolRegistry + confirmation and safety gates.Phase 5: migrated all 50 tools to registry-attached handlers.Phase 6: unified memory and remaining LLM consumers around the canonical architecture.The project currently has 1000+ automated Python tests, live end-to-end scenarios, and security regression tests.🔐 Safety is a major part of FOLFOL isn’t supposed to blindly execute everything an LLM asks for.High-risk actions such as clicking, typing, sending messages, shell commands, Safari JavaScript, and cookie operations can require explicit user confirmation.The confirmation system is enforced by code, not by the model itself.Unknown tools fail closed, approvals are bound to the exact tool-call arguments, and secrets are scrubbed before being written to memory.🎯 What I’m working on nowI’m continuing the architecture migration and working toward a more unified FOL core where:Memory + LLM + Tools + Agents + UI + Safetywork together as one coherent system.The long-term goal is simple:Build a personal AI that doesn’t just chat with me, but can actually understand my environment, remember context, use tools, and help me operate my computer.This is still a work in progress, but FOL is becoming much more than a chatbot. 🚀#Stardance #HackClub #AI #macOS #Python #SwiftUI #LLM #AIAssistant #OpenSource