FloatGPT — Omnipresent, Local-First AI Desktop Companion

FloatGPT is an omnipresent desktop AI companion that floats over Windows to automate OS workflows, plan tasks, and generate real-time executive dossiers. Operating 100% privately on your PC or powered by the cloud, it transforms daily computing with screen awareness and calm execution.

Live demoGitHub1

Overview

# 🪐 FloatGPT — The Omnipresent Desktop AI Coworker

### Calm execution, deep OS orchestration, and real-time executive intelligence—floating right on your desktop.

---

## 💡 The Inspiration

Every current AI assistant suffers from the same fundamental flaw: **it is trapped in a browser tab.**

Whenever you need help while coding, analyzing documents, or attending meetings, you are forced into an exhausting loop of friction: switching windows, copying text, pasting it into a chatbox, copying the response, and pasting it back into your work.

We built **FloatGPT** to solve this. Instead of making you come to the AI, FloatGPT lives as an **omnipresent, floating desktop Orb** that hovers quietly above your Windows workspace, ready to execute tasks, analyze your screen, and orchestrate deep project roadmaps with zero context switching.

---

## ✨ What FloatGPT Does

FloatGPT is a local-first, floating AI desktop companion engineered around the philosophy of **"Calm Execution"**:

### 1. 👔 PreBrief — 30-Day Executive Meeting Intelligence

- **Temporal Web Intelligence:** Leverages Tavily's 30-day temporal research streams to gather fresh, real-time news, executive movements, and financial developments.

- **Debris & SEO Firewall:** Filters out crawler clutter and sponsored fluff to extract key decision makers, corporate vitals, and tailored icebreakers.

- **1-Click PDF Dossiers:** Exports print-ready, multi-block executive briefing dossiers rendered with native dark-mode canvas styling directly into your Flowtion shelf.

### 2. 🎯 Dual Modes: Chat vs. Plan Execution

- **Chat Mode:** Rapid-fire Q&A, voice push-to-talk, screen glance awareness, and quick slash commands.

- **Plan Mode:** Turns any ambiguous objective into ordered, trackable execution steps with deadlines and 1-click progress verification.

- **Persistent Context:** Define your custom working persona, guidelines, and project rules in the Context drawer.

### 3. 🏠 Home Hub & Productivity Utilities

- **FlowPing WhatsApp:** Schedule messages with exact clock times and international numbers (+91 / global) that dispatch reliably in the background.

- **Price Compare:** Real-time multi-retailer price aggregator with verified links and discount rankings.

- **Flowtion Local Shelf:** Store markdown notes, research papers, and PDF documents safely on-device.

- **PC Doctor & Digital Guardian:** Real-time CPU/RAM telemetry, process diagnostics, and focus enforcement blocklists that keep you in flow.

### 4. 🎨 Dual Studio Surfaces: Calm & Flow

- **Flow Mode (Default):** A tactile, warm cream-paper workspace (#f3eee2) with dot-grid styling and editorial typography.

- **Calm Mode:** Distraction-free obsidian glass with subtle neon accents designed for deep coding and late-night focus.

---

## 🔒 Local-First Privacy Meets Cloud Flexibility

FloatGPT gives users complete data sovereignty:

- **100% Offline with Local Models:** Pull models via Ollama (Qwen 2.5 Coder 7B, DeepSeek, GPT-OSS) and run FloatGPT completely disconnected from the internet.

- **Cloud Supercharge:** Plug in your own API keys for Google Gemini, Groq (ultra-fast Llama), or OpenAI for heavy reasoning and vision tasks.

- **Zero-Data Leakage:** Memory, transcripts, and personal settings stay stored locally on your PC via encrypted storage.

---

## 🛠️ How We Built It

- **Frontend & Architecture:** React 19, Vite, Tailwind CSS v4, and Framer Motion for physics-based layout transitions.

- **Desktop Core:** Electron with native Win32 window-dragging mathematical constraints bound to screen DPI.

- **Routing Engine:** FastRouter for zero-token deterministic OS actions (system clock, calculator, battery diag) before escalating to LLM inference.

- **OS Execution & Safety Layer:** PowerShell automation via orchestrator.ts protected by a strict scriptFirewall interceptor that blocks destructive operations (rm -rf, format, del).

- **Document & PDF Engine:** Custom pdfEngine with dark-mode canvas rendering and real-time text extraction.

---

## 🔮 What's Next for FloatGPT

- **Universal Model Context Protocol (MCP) Host:** Connecting the floating Orb to external tools (GitHub, Notion, PostgreSQL, and Google Workspace).

- **1-Click Clean Downloads:** Context-aware automatic file renamer that turns messy Downloads into organized, searchable folders without moving file paths.

- **Hardware Trackpad Sliders:** Using low-level Precision Touchpad hooks to turn physical laptop trackpads into volume and brightness slider decks.

How it uses Vultr

FloatGPT is engineered as an ultra-fast, local-first ambient intelligence operating directly on user desktops. However, ensuring universal access across everyday hardware—especially 8GB laptops—demanded a high-throughput, low-latency cloud intelligence tier. In our current architecture and next-gen releases, Vultr serves as the foundational backbone powering FloatGPT's distributed intelligence, computation pipelines, and multi-domain synchronization. 1. Cloud GPU Clusters (Inference & Routing): While power users run quantized models locally, everyday mobile users lack dedicated VRAM. We leverage Vultr Cloud GPUs (NVIDIA A100/L40S) orchestrated via vLLM. This cluster delivers sub-120ms token latency for our dual-tier engine, dynamically routing complex OS orchestration, live code synthesis, and multi-step reasoning without local throttling. 2. Compute Instances (PreBrief & Heavy Tasks): Intensive background jobs—such as our PreBrief research engine, deep web crawling, DOM debris sanitization, and heavy PDF compiling—are decoupled from client machines and routed to dedicated Vultr Compute instances. This ensures user devices remain cool and responsive during heavy data ingestion. 3. Vultr S3-Compatible Object Storage: User privacy is paramount. FloatGPT stores zero telemetry; encrypted Flowtion markdown vaults, vector snapshots, and exported media sync seamlessly to Vultr Object Storage. S3 compatibility enabled rapid deployment with end-to-end client encryption. 4. Global Edge Routing & Future Initiatives: As we expand our domain ecosystem, Vultr's multi-region footprint acts as our intelligent edge gateway, minimizing latency worldwide. Vultr provides the high-bandwidth, cost-effective infrastructure that makes enterprise AI instantly accessible on any desktop.

1 comment

Sorted by votes

Really interesting project !