Skip to content

ZANYANBU/Chrome_Assist_AI

BranchesTags

Repository files navigation

ZANYSURF - Autonomous AI Browser Agent

Give it a goal. It does the work.

Manifest Chrome Edge Add-ons Works Offline Version License

ZANYSURF is a Chrome and Edge extension that turns any LLM into a private, autonomous web agent. Type what you want in plain English. It plans, browses, clicks, fills forms, and reports back with full transparency.

If ZANYSURF saves you time, please star the repo.

🎬 See It In Action

ZANYSURF Demo

Demo GIF coming soon — star the repo to get notified!

Highlights

  • Summarize YouTube: extract key takeaways from long videos and jump to timestamps.
  • All-in-one models: OpenAI, Claude 4, Gemini, Llama, and more (bring your own keys).
  • Chat with PDFs and pages: drop PDF, DOC, TXT, or XLS to get answers fast.
  • Dive the web: research any site with answers and citations.
  • Response faster: set tone and craft emails, replies, or tweets in seconds.
  • Monitor prices: track price drops and back-in-stock alerts across marketplaces.
  • Automate your work: navigate, extract, click, and fill forms.
  • Integrate with 1000+ apps: Make.com and Zapier workflows.
  • Record browser macros and replay them instantly.
  • REST API surface for external automation and CI/CD pipelines.

Quick Start

Option A - Free, local (recommended)

  1. Install Ollama: https://ollama.com
  2. Run: ollama pull llama3.2
  3. Load the extension/ folder in Chrome or Edge (developer mode)
  4. Open the ZANYSURF side panel, select Ollama, type your goal

Option B - Cloud API

  1. Load the extension/ folder
  2. Settings -> choose provider -> add API key
  3. Type your goal and go

Option C - Edge built-in AI (if available)

  1. Load the extension/ folder in Edge
  2. Select Edge Built-in AI

Screenshots

Add screenshots here:

  • docs/screenshots/overview.png
  • docs/screenshots/agent-run.png
  • docs/screenshots/price-compare.png
  • docs/screenshots/settings.png

Example usage in README:

Overview Agent Run

See the screenshot placeholder guide: docs/screenshots/README.md

Features

Core agent

  • Plan-and-execute with reflexion
  • Multi-tab orchestration with dependency graphs
  • Vision mode for sparse DOM pages
  • Safe mode with approval gates
  • Local memory and knowledge graph

DOM engine (v2.4.0)

  • MutationObserver-based DOM stability detection (replaces polling)
  • Shadow DOM and SPA hydration awareness
  • React fiber idle check before capturing DOM
  • Exponential-backoff retry on transient element failures (stale, detached)
  • Stable for 400 ms window before agent reads the page
  • Top-level error boundary in content script message handler

Macro Recorder

  • Record any sequence of browser actions into a named macro
  • Macros are stored locally (100 saved, no cloud)
  • Replay a macro on any active tab with a single command
  • Delete macros and inspect recorded steps
  • Integrate macros into workflows or trigger via REST API

REST API

  • External messaging surface via chrome.runtime.sendMessage
  • Run agent, stop agent, get live status, get metrics
  • List and replay workflows and macros
  • Read or clear memory context
  • Enqueue parallel task batches
  • Retrieve audit log and API cost metrics

Automation

  • Scheduler for recurring goals
  • Workflow replay and audit logs
  • CSV export for extracted data

Price comparison

  • Auto-open marketplace tabs
  • Extract prices per tab
  • Synthesize results and export CSV

Macro Recorder

Record any sequence of browser actions and replay them later — no code required.

Start recording

  1. Open the ZANYSURF side panel
  2. Click Record Macro (or send START_MACRO_RECORDING via the extension API)
  3. Perform your actions — clicks, form fills, navigation
  4. Click Stop to save the macro with a name

Replay

  • Select a saved macro from the list and click Replay
  • Or trigger via the REST API: REPLAY_MACRO with { macroId }

Via extension message API

// Start recording
chrome.runtime.sendMessage(EXTENSION_ID, { action: 'START_MACRO_RECORDING', goal: 'Login flow' });

// Stop and get steps
chrome.runtime.sendMessage(EXTENSION_ID, { action: 'STOP_MACRO_RECORDING' });

// Save
chrome.runtime.sendMessage(EXTENSION_ID, { action: 'SAVE_MACRO', name: 'Login flow' });

// Replay by ID
chrome.runtime.sendMessage(EXTENSION_ID, { action: 'REPLAY_MACRO', macroId: '<id>' });

// List all macros
chrome.runtime.sendMessage(EXTENSION_ID, { action: 'LIST_MACROS' });

// Delete
chrome.runtime.sendMessage(EXTENSION_ID, { action: 'DELETE_MACRO', macroId: '<id>' });

REST API

ZANYSURF exposes an external messaging API that any extension (or native messaging bridge) can call.

Connection: use chrome.runtime.sendMessage(ZANYSURF_EXTENSION_ID, { action, ...params }).

The extension ID must be added to your caller extension's externally_connectable and to ZANYSURF's manifest.json externally_connectable.matches.

Action Params Response
RUN_AGENT { goal } { success, result }
STOP_AGENT — { success }
GET_STATUS — { active, goal, steps }
GET_AGENT_METRICS — { success, metrics }
GET_WORKFLOWS — { success, workflows[] }
REPLAY_WORKFLOW { workflowId } { success, result }
GET_MACROS — { success, macros[] }
SAVE_MACRO { name, steps[] } { success, macro }
REPLAY_MACRO { macroId } { success, result }
DELETE_MACRO { macroId } { success }
GET_MEMORY { query? } { success, memory[] }
CLEAR_MEMORY — { success }
ENQUEUE_TASKS { tasks[] } { success, queued }
GET_TASK_STATUS — { success, snapshot }
GET_AUDIT_LOG — { success, log[] }
GET_API_METRICS — { success, metrics }

Example — run agent from another extension

const ZANYSURF_ID = '<extension-id>';
chrome.runtime.sendMessage(ZANYSURF_ID, {
  action: 'RUN_AGENT',
  goal: 'Search for "best espresso machine 2026" and return top 3 results'
}, response => {
  console.log(response.result);
});
Provider Notes
Ollama Local and private, no server required
Gemini Long context for research tasks
OpenAI General purpose
Claude Strong reasoning
Groq Very fast
Mistral Cost efficient
Edge Built-in Zero-setup on Edge

Install (Chrome and Edge)

⚠️ API Key Notice: If using Gemini, your API key is stored in chrome.storage.local. It never leaves your browser but can be accessed via Chrome DevTools. Use a key with usage limits set at Google AI Studio.

Chrome

  1. Open chrome://extensions
  2. Enable Developer mode
  3. Load unpacked -> select extension/

Edge

  1. Open edge://extensions
  2. Enable Developer mode
  3. Load unpacked -> select extension/

From source

git clone https://github.com/ZANYANBU/Chrome_Assist_AI.git
cd Chrome_Assist_AI
npm install
npm run build

Configuration

Open the ZANYSURF side panel and click the settings icon.

Setting Description
Provider Choose Ollama, Gemini, OpenAI, Claude, Groq, Mistral, or Edge Built-in
Ollama URL Default: http://localhost:11434
API Key Encrypted in the local vault
Safe Mode Require approval for risky actions
Memory Toggle short-term and long-term memory

Permissions Explained

Permission Why it is needed
activeTab Read and interact with the current page
scripting Inject scripts for actions
storage Save settings, memory, and vault
alarms Run scheduled tasks
tabs Multi-tab orchestration
downloads CSV exports
sidePanel Persistent UI in Chrome/Edge

Architecture (High Level)

Popup and side panel -> Service worker (agent loop) -> Content script (DOM + actions)

Key components:

  • LLMGateway: provider routing
  • MemorySystem: short/long memory + retrieval
  • OrchestratorAgent: multi-agent pipelines
  • Risk guards: approvals for critical actions

Changelog

Mar 5, 2026 — v2.4.0

  • DOM fixes: replaced interval polling with MutationObserver-based waitForDomStable. Now detects React/Vue/Angular hydration correctly via shadow DOM observation and React fiber idle check.
  • Stability fixes: added exponential-backoff retry (200ms/400ms) for transient element failures; top-level try/catch in content script message handler prevents one bad handler from crashing the rest; removed duplicate sleep declaration.
  • Macro Recorder: full record, save, replay, delete pipeline. Steps captured from content script, stored in chrome.storage.local, replayed tab-by-tab with proper wait logic.
  • REST API: expanded onMessageExternal with 16 endpoints covering agent control, workflows, macros, memory, tasks, and audit log.

Mar 4, 2026 — v2.1.0

  • Added price comparison planning and marketplace search URLs.
  • Improved vision-mode click reliability.
  • Reduced prompt bloat with context and DOM budgeting.
  • Added chat vs task intent detection.

Docs

Repository Structure

ZANYSURF/
├── extension/              ← load this folder in Chrome / Edge (developer mode)
│   ├── background.js
│   ├── content.js
│   ├── dom-text-worker.js
│   ├── popup.html / popup.js / popup.css
│   ├── selftest.html / selftest.js
│   ├── manifest.json
│   ├── manifest.edge.json
│   ├── metadata.json
│   └── icons/
├── dev-playground/         ← Vite + React build tooling (NOT the extension)
│   ├── README.md           ← explains this folder
│   └── App.tsx, vite.config.ts, package.json …
├── docs/
│   ├── demo.gif            ← demo screencast (coming soon)
│   ├── LAUNCH.md
│   └── screenshots/
├── qa/                     ← smoke tests and QA reports
├── launch/                 ← store listing and launch assets
├── .github/                ← issue templates and CI workflow
├── .gitignore
├── README.md
├── CHANGELOG.md
├── PRIVACY.md
├── CONTRIBUTING.md
├── PERMISSIONS.md
└── LICENSE

Contributing

  1. Fork the repo
  2. Create a feature branch: git checkout -b feature/your-feature
  3. Commit: git commit -m "feat: your feature"
  4. Push and open a Pull Request

License

MIT License - see LICENSE.

Privacy policy: PRIVACY.md

About

Autonomous AI browser agent — 6 LLM providers, runs locally with Ollama, async multi-tab, memory decay. Chrome & Edge.

Topics

chrome-extension ai-agents autonomous-agent manifest-v3 browser-agent ollama local-aiprivacy

Resources

Readme

License

MIT license

Contributing

Contributing
Activity

Stars

1 star

Watchers

0 watching

Forks

0 forks
Report repository

Releases

2 tags

Packages

 
 
 

Contributors

Languages