nikola66

nikola66 / web-agent

Public

Browser-native agent · profiles · tools — by aratech | Zero installs, isolated, secured and self-evolving agent.

webagent.aratech.ae agentic-ai hermes-agent openclaw self-improving-agent self-improving-ai
11
1
85% credibility
Found May 17, 2026 at 11 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
TypeScript
AI Summary

Web Agent is an open-source AI assistant that runs entirely inside your web browser. Unlike traditional AI tools that require servers or installations, Web Agent uses WebContainers technology to create a complete Node.js runtime right in your browser tab. You create isolated profiles with their own workspaces, memory, and settings—all stored locally on your machine. The assistant can read and write files, search the web, send emails, manage reusable skills, and run scheduled tasks. It features special commands like /plan for drafting detailed plans before execution, /wiki-setup for organizing knowledge in a personal vault, and /skills for saving reusable workflows. Everything is designed to feel simple for everyday users while offering powerful capabilities for developers. Your data never leaves your browser unless you explicitly ask it to.

How It Works

1
🔍 You discover Web Agent

You hear about a smart assistant that lives entirely in your browser—no downloads, no servers, no complicated setup.

2
🌐 You open the app

You visit the web address and see a clean, modern interface with a terminal-style chat window ready to go.

3
👤 You create your profile

You give your assistant a name and personality. Everything stays private in your browser—no accounts needed.

4
🔌 You connect your AI service

You paste in a free API key from OpenRouter or point to your own AI provider. Your assistant is now powered up and ready to think.

5
🚀 You launch your assistant

With one click, your assistant boots up inside a secure workspace. It greets you warmly and is ready to help.

6
You start chatting
📁
Work with files

Ask it to write code, organize projects, or explore folders in its workspace

🧠
Remember things

Save facts, session notes, and learnings that improve over time

📋
Make plans

Use /plan to research and draft a detailed plan before starting work

✨ You get things done

Your assistant completes tasks, remembers your preferences, and keeps improving. Everything stays on your computer—no cloud dependency.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 11 to 11 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is web-agent?

web-agent is a browser-native AI agent that runs entirely in your browser—no Docker, no VPS, no local Python stack. It leverages WebContainers to spin up a Node.js runtime directly in the browser tab, giving you an isolated workspace with persistent memory, built-in tools, and a self-improving agent that learns over time.

You open the app, connect an API key (OpenRouter, Ollama cloud, or any OpenAI-compatible endpoint), and start chatting. The agent can read and write files, run shell commands, search the web, analyze images, send emails, and maintain a knowledge vault using PARA-style markdown structure synced to Obsidian if you want it.

Why is it gaining traction?

The pitch is simplicity: zero setup, maximum capability. Most AI agent tools require infrastructure you have to maintain. web-agent sidesteps that entirely—it's a web app you can host anywhere, while user state stays browser-local and encrypted.

The `/plan` workflow is genuinely useful for developer workflows. You describe a goal, the agent researches your workspace with read-only tools, writes a timestamped markdown plan, and waits for your approval before executing. It's spec-first without the ceremony.

The knowledge vault (`wiki_*` tools) projects facts, session notes, and learnings into a browseable markdown tree. If you live in Obsidian or similar, this bridges the gap between "agent memory" and "human-readable notes" without duplicating work.

Who should use this?

Developers who want to experiment with AI agents without infrastructure commitment. If you've wanted to test an agent workflow but balked at the Docker/VPS overhead, this is a legitimate option. It's also useful for evaluating AI agent designs—the isolated profiles let you run parallel experiments with different models or personalities.

Note: With only 11 stars, this is early-stage software. The credibility score reflects that maturity level. The feature surface is broad, but you should test thoroughly before relying on it for production workflows.

Verdict

web-agent solves the "I want to try an AI agent but don't want to set up a server" problem with a clever browser-native architecture. The feature depth—planning modes, knowledge vaults, skills as reusable markdown, multi-provider support—is impressive for a project this size.

However, the 11-star count signals a project in its infancy. The codebase appears well-structured and the documentation is thorough, but evaluate it as a proof-of-concept worth watching rather than a production-ready tool. Try the hosted demo at webagent.aratech.ae before committing dev time.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.