FlowForge

Forging your experience...

FlowForge is a Web Onboarding Assistant that turns user intent into actionable UI guidance inside web applications.

• Ask questions about the current page
• Find relevant UI elements
• Get step-by-step onboarding flows

Demo

Try the live Demo at useflowforge.app ✦

Overview

Modern web apps are powerful but hard to navigate. The Assistant lets you ask natural-language questions and get contextual guidance directly in the UI.

It can:

• highlight relevant elements
• explain what's on the page
• guide users step-by-step through workflows

Under the hood: browser extension + AI agent (ReAct) + RAG pipeline.

Use cases

• Ask a question and get a clear explanation of what the page is about.
• Locate links, buttons, or contacts without manual search.
• Follow step-by-step actions directly in the interface.
• Highlight and summarize important parts of long pages.
• Understand unfamiliar interfaces without onboarding.

Disclaimer

This is an early-stage experimental project showcasing the ✦ Idea ✦ itself.

It is not production-ready, and the core engine is still evolving. Expect limitations, rough edges, and ongoing improvements.

Changelog – see the list of changes.

Quick start

Prerequisites

• Node.js 22+
• Chrome / Chromium browser
• Ollama (optional, for local LLM)

Run

npm i
npm start

This installs dependencies, builds backend and extension, guides Chrome installation, and starts the backend at http://localhost:3477.

Usage

1. Open any website
2. Click the Assistant extension icon
3. Ask a question, e.g.:
   • "Where is the login button?"
   • "How do I checkout?"
   • "Show me the search bar"
4. View the answer and highlighted elements directly on the page

Security

• DOM-only access — operates strictly on the visible page DOM, without access to cookies or session storage
• Sensitive data filtering — excludes inputs like passwords, payment details, and other private fields during extraction
• Local-first architecture — backend and vector storage run locally by default
• Optional local AI — supports fully local inference via Ollama (no external API required)

Note: As an early-stage MVP, security is evolving and not yet production-ready.

Roadmap

Focus areas:

1. Improve the core engine: extractors, embeddings, reasoning, and tools
2. Extend context from single pages to full websites
3. Build a proper standalone extension distribution

Backlog — the full list of features and improvements

Documentation

• Architecture — system design and component interaction
• DOM to RAG pipeline – from DOM to actionable intelligence
• PageTrail — DOM representation for UI understanding
• Backend — AI agent backend and RAG pipeline
• Extension — browser extension and page extraction

License

MIT License ✦ 2026 Antony Belov

See LICENSE for details.