Welcome to Loopi

Loopi is the open‑source, local‑first automation platform that combines a visual builder, real browser control, desktop cursor & keyboard, shell commands, 80+ API integrations, and self‑patching AI agents — all running locally on your machine.

What is Loopi?

Loopi provides a visual node‑based editor for building automations that can:

• Drive the browser — navigate, click, type, extract, upload, screenshot, headless or visible.
• Control the desktop — move the cursor, click/drag, scroll, type text, fire hotkeys, take screenshots. X11, Wayland, macOS, Windows.
• Run shell commands — capture stdout/stderr/exit code into typed variables; pipe between steps.
• Connect to 80+ services — Slack, Discord, GitHub, Notion, Stripe, Postgres, SendGrid, and more.
• Use any LLM — OpenAI, Anthropic, Ollama, or the Claude Code CLI. Pick per‑agent.
• Transform data — JSON parse, math, string ops, array filter/map, date/time, inline JavaScript.
• Use typed variables — auto‑detected types with dot notation and array indexing.
• Branch and loop — conditional logic with if/else and forEach iteration.
• Store credentials — encrypted storage (AES‑256‑CBC) for API keys, tokens, auth.
• Schedule runs — interval, cron expressions, or one‑time.
• Run agents — goal‑driven workers that reflect on each run and rewrite failing workflows in place.

Quick Links

AI Agents — New in v1.9

Agents bundle a goal, a set of workflows, a schedule, and a model. After every run, the reflection engine reads per‑node outcomes and the final variables, compares them to the goal, and picks a verdict:

• ok — the run made progress; leave the workflow alone.
• modify — the workflow is structurally wrong; rewrite it in place with a validated replacement graph.
• fail — the run failed in a way reflection can't repair.

Reflection is model‑gated: small models (like Claude Haiku or GPT‑3.5‑turbo) are blocked at creation time because they silently return ok without really engaging.

See the Agents guide for details.

AI Providers

Connect local or cloud LLMs to your workflows:

Anthropic

Claude Sonnet 4.5, Opus 4 — best all‑rounders for reflection and long‑context reasoning.

OpenAI

GPT‑4o, GPT‑4o‑mini, GPT‑4‑turbo for fast cloud reasoning.

Claude Code

Uses the Claude Code CLI that's already installed on your machine. No API key. Zero config.

Ollama

Llama, Mistral, Mixtral, Qwen, DeepSeek — runs 100% locally. Your data never leaves the machine.

80+ Built-In Integrations

No plugins. All integrations are included:

Each integration supports credential selection via dropdown — store your API keys in Settings and pick them when configuring steps.

Key Features

Visual node-based editor

Drag‑and‑drop interface powered by ReactFlow. No coding required. Typed variables, conditionals, forEach loops.

Real browser control

Electron BrowserWindow + webContents automation. Real Chromium window — navigate, click, type, extract, screenshot. Watch it live or run fully headless.

Desktop cursor & keyboard

Move the mouse, click (left/right/double), drag, scroll, type text, press keys, fire hotkey combos (Ctrl+C, Alt+Tab, …) anywhere on the desktop. Full support for Linux Wayland via ydotool, X11 via nut-js, native support on macOS and Windows.

Shell commands as steps

Run any shell command from a workflow. Capture stdout, stderr, and exit code into typed variables. Set working directory, timeout, shell. Pipe output into subsequent steps via variable substitution.

Desktop screenshots

Capture the full screen or specific regions. Works on Wayland (via Electron desktopCapturer), X11 (via nut-js), and all platforms. Save to custom paths, store paths in variables.

Auto-typed variable system

Variables automatically detect types and support full access syntax:

• Simple: {{username}}
• Nested: {{user.name}}
• Arrays: {{users[0]}}
• Mixed: {{users[0].profile.email}}
• Deep: {{response.data[0].meta.created_at}}

Credentials manager

Encrypted storage (AES‑256‑CBC, PBKDF2 key derivation) for 80+ credential types. Select credentials from a dropdown when configuring integration steps.

Conditional logic & loops

• Browser conditionals — branch based on element existence or extracted values.
• Variable conditionals — branch based on variable comparisons.
• ForEach loops — iterate over arrays with loop and done edges.

Scheduling

Interval, cron, or one‑time. Desktop notifications on start/finish/error.

CLI

Run and manage workflows from your terminal with loopi-cli. Shares credentials and browser sessions with the desktop app. See the CLI guide.

How Loopi Is Different

• vs n8n — Real browser control, desktop input, and local AI. Not just API orchestration.
• vs Zapier/Make — Local‑first with free local AI via Ollama. No cloud lock‑in.
• vs Playwright/Puppeteer — Visual builder with 80+ integrations and AI. No coding.
• vs Selenium IDE — Typed variables, data transforms, AI, and API nodes built in.
• vs RPA suites — Open source, lightweight, free. No vendor lock‑in.
• vs AI agent frameworks — Visual builder — see your agents run in a real browser, with self‑patching workflows.

Getting Help

• GitHub Issues — Report bugs or request features
• Discussions — Ask questions and share workflows
• Documentation — Browse the guides in this documentation

Contributing

We welcome contributions — especially around agent capabilities, new integrations, desktop primitives, and workflow templates:

• Add new AI model providers (Gemini, Cohere, local GGUF models)
• Build agent templates (research, monitoring, content pipelines)
• Add new service integrations
• Improve the AI copilot (auto‑generate workflows from natural language)
• Add tool‑use / function‑calling support for connected LLMs

Ready to get started? Head to the Getting Started guide, then check out Agents.