I built an open-source tool that gives your local AI model a real browser and 80+ APIs — no code needed

Loopi is an open-source desktop automation platform I've been building to solve a problem I kept running into: most automation tools make you choose between visual (but cloud-only) or powerful (but code-only). Why not both?

The Problem

Here's what a typical "automate something with AI" project looks like today:

• Zapier/Make for API orchestration — but cloud-only, no browser control, expensive at scale
• Playwright/Puppeteer for browser automation — but requires writing and maintaining code
• LangChain/AutoGPT for AI agents — but entirely code-based, hard to debug, no visual feedback

Three tools. Three codebases. Three sets of credentials. Constant context switching.

What Loopi Does

Loopi combines all three into a single visual desktop app:

• 🖱️ Visual drag-and-drop workflow builder — powered by ReactFlow, no code required
• 🌐 Real Chromium browser control — navigate, click, type, extract, screenshot in a real browser window you can watch live
• 🤖 Local + cloud AI — connect OpenAI, Anthropic Claude, or run Llama/Mistral/Gemma locally with Ollama (your data never leaves your machine)
• 🔌 80+ built-in integrations — Slack, Discord, GitHub, Notion, Postgres, MongoDB, Stripe, Gmail, Telegram, S3, and many more
• 📊 Typed variable system — {{user.name}}, {{items[0].price}}, auto-detected types with dot notation
• ⏰ Scheduling — run on intervals, cron expressions, or one-time schedules

Building Your First AI Agent

Here's what a real workflow looks like in Loopi:

[Schedule: every morning 8am]
    ↓
[Browser: navigate to competitor pricing page]
    ↓
[Browser: extract text from .pricing-table]
    ↓
[Ollama/Llama: "Summarize any price changes vs last week: {{extracted_text}}"]
    ↓
[Slack: post summary to #competitive-intel]

That entire workflow is built by dragging nodes — no code. The Ollama node runs locally, so the competitor data never hits a third-party server.

Why Local AI Matters

The Ollama integration is the one I'm most proud of. When you connect a local model:

• No API keys needed
• No per-token costs
• Your workflow data stays on your machine
• Works completely offline
• Supports Llama 3, Mistral, Gemma, Phi, CodeLlama, and any other Ollama-compatible model

For privacy-sensitive automations (processing emails, internal documents, customer data), this is huge.

How It Compares

Feature	Loopi	n8n	Zapier	Playwright
Visual builder	✅	✅	✅	❌
Real browser control	✅	❌	❌	✅
Local AI (Ollama)	✅	❌	❌	❌
Runs locally	✅	Self-host	❌	✅
Open source	✅	✅	❌	✅
Free	✅	Limits	❌	✅

Loopi is the only tool that gives you all of these together.

The Tech Stack

For those who want to contribute or just geek out:

• Electron — cross-platform desktop app
• React 19 — UI
• TypeScript — throughout
• ReactFlow — the node editor
• Tailwind CSS + Radix UI — styling
• Biome — formatting and linting

What We're Building Next

The areas we most want help with:

• New AI model providers — Gemini, Cohere, local GGUF models
• Agentic workflow templates — research agents, monitoring agents, content pipelines
• New service integrations — the 80+ is a start, not a finish
• AI copilot improvements — auto-generate workflows from natural language descriptions
• Tool-use / function-calling — for connected LLMs

Getting Started

# Download the latest release
# Windows: .exe installer
# Linux: .deb package

# Or build from source:
git clone https://github.com/Dyan-Dev/loopi.git
cd loopi
pnpm install
pnpm start

• 🌐 Website: https://loopi.dyan.live/
• 📁 GitHub: https://github.com/Dyan-Dev/loopi
• 🎥 Demo video: https://youtu.be/QLP-VOGVHBc

The project has 144 GitHub stars and is growing. If you're into automation, AI agents, or just want to give your local LLM actual hands and eyes — give it a try and let me know what you think.

PRs and ideas are very welcome 🙏