_{OpenBUA was asked to "Find me 2 students from Australia using Linkedin and email them selling my app Petedoro"}

---

Claude Code gave AI agents access to people's terminals. OpenBUA gives agents access to the one place users actually live all day - their own browser, already logged into Gmail, LinkedIn, GitHub, job boards and CRMs.

---

OpenBUA (Open Browser Use Agent) is an open-source, autonomous AI browser agent built as a Chrome Extension (Manifest V3) that runs directly inside your everyday browser. Unlike cloud browser-use tools that require spun-up headless containers, login bypass proxies, or remote servers, OpenBUA operates right inside your existing browser session with all your active cookies, credentials, and logins already available.

Built with @earendil-works/pi-agent-core and @earendil-works/pi-ai, OpenBUA equips AI models with real-time DOM perception, intelligent multi-step navigation, synthetic event form-filling, incremental live research previews, and a self-learning memory system.

---

Why OpenBUA? In-Browser Execution vs Cloud Browser-Use

Capability	OpenBUA	Traditional Cloud Browser-Use
Session Authentication	Uses your active, signed-in sessions (Gmail, LinkedIn, GitHub, CRM)	Requires re-authentication, 2FA bypass, or fragile cookie imports
Infrastructure	100% Client-Side Chrome Extension (Zero servers)	Spun-up Docker containers, cloud VMs, remote Playwright instances
Data Privacy	Zero data leaves your machine; prompts go straight to AI provider	Browsing traffic and page snapshots stream through remote servers
Cost	Free Gemini tier (1,500 requests/day) or Bring Your Own Key	Per-minute cloud compute pricing plus proxy bandwidth costs
Bot Detection	Bypasses headless blocks by running as a genuine local browser	Regularly blocked by Cloudflare, reCAPTCHA, and bot shields
Captcha Handling	Human-in-the-Loop alert banner lets you solve challenges in 1 click	Requires expensive third-party captcha-solving API integrations
Real-Time Feedback	Live Research Preview tab streams tables incrementally as agent works	Black-box execution until final summary is returned
Knowledge Retention	Proactive self-learning memory with Global and Tab memory tiers	Ephemeral scratchpads discarded after each session

---

What Can OpenBUA Do?

Because OpenBUA runs directly in your existing browser, it executes end-to-end autonomous workflows without login barriers:

• Autonomous Web Research and Lead Generation:

  • Search directories, platforms, search engines, or social networks across multiple pages.
  • Automatically evaluate candidates, extract verified data, and compile structured tables.
  • Example: "Search LinkedIn for 'world models PhD' researchers, extract profiles across pages, and compile a structured table with names, headlines, and locations."
  • Example: "Go through recent Hacker News submissions on LLM agents, find the top 5 discussions, and summarize key takeaways."

• Live Research Preview Tab:

  • Dedicated Preview tab accessible from the center navigation toggle (Chat / Memory / Preview).
  • Incremental write-as-you-go streaming: The agent streams extracted candidates, table rows, and structured Markdown into the preview buffer as each item is inspected (ReAct 1-item cycle), eliminating black-box waiting.
  • Automatic Markdown table repair engine: Automatically detects and stitches fragmented rows and scattered pipe data into contiguous tables.
  • Scoped per chat session: Each conversation tab maintains its own independent preview buffer with an instant Markdown copy button.

• Self-Learning Memory System:

  • Proactive fact extraction: Detects personal details, contact information, university affiliations, and repeatable workflow routines while browsing.
  • Suggested Memories interface: Review suggested memories on a dedicated full-page screen, discard with a single tap, or approve into Tab Memory or Global Memory.
  • Global Memory: General user background (profile, resume, work history) shared across all chats and browser tabs.
  • Tab Memory: Project-specific notes and uploaded documents isolated to the active chat session.
  • Document Uploads: Directly upload PDF, Markdown, Text, JSON, image, and video files with client-side text parsing powered by pdfjs-dist.
  • Large Media Storage: Videos and files over 3 MB are kept as raw Blobs in IndexedDB instead of base64 strings in chrome.storage.local, so uploading a long video no longer freezes the panel. Inline dataUrl is kept only for small files that need OCR.

• Cross-Platform Social Media Posting:

  • One post_to_social call posts media + caption to X/Twitter, LinkedIn, Reddit, Facebook, Instagram, Threads, Bluesky, Mastodon, YouTube, Pinterest, Tumblr, or TikTok.
  • Resolves the platform from your request or the current tab, opens its composer, attaches stored videos/images, types the caption, and reports composer state so nothing publishes without your say-so.
  • Files of any size are streamed to the page in bounded chunks, so a 200 MB video attaches exactly like a small image.
  • Correctly targets the file input inside the open composer dialog, which is what single-page social apps require.

• Autonomous Form Filling and Wizard Completion:

  • Matches web forms, job application portals (Greenhouse, Lever, Workday), and customer onboarding wizards against your stored profile.
  • Dispatches native synthetic event chains (input, change, blur) ensuring 100% compatibility with React, Vue, Angular, Svelte, and vanilla HTML forms.
  • Verifies DOM values after population to prevent hallucinated submissions.

• Autonomous Email and Webmail Outreach (Gmail, Outlook, Webmail):

  • Drafts, reviews, and dispatches genuine emails right from your existing, logged-in browser session (Gmail, Outlook, Yahoo, webmail). No third-party OAuth permissions, SMTP passwords, or external email APIs required.
  • Navigates to your webmail interface, triggers compose dialogs, inputs confirmed recipient chips, formats subject lines and body text, and clicks Send on your command.
  • Features intelligent Gmail Basic HTML mode navigation fallback to eliminate DOM drift and click friction on dynamic Single-Page Applications.

• Multi-Tab Orchestration and Dynamic Automation:

  • Opens, switches, and coordinates across browser tabs during multi-step workflows without losing current task progress or form context.
  • Paginated scraping, scrolling dynamically to trigger lazy-loaded elements, and interacting with buttons, dropdowns, and dialogs.

• Human-in-the-Loop (HITL) CAPTCHA Handling:

  • Detects bot verification challenges (Cloudflare Turnstile, Google reCAPTCHA, hCaptcha, Arkose Labs) on visited sites.
  • Displays a clean in-panel alert allowing you to solve the challenge manually, then automatically resumes automation upon completion.
  • Employs intelligent fallbacks and exploration budgets to prevent infinite navigation loops.

• Chain-of-Thought Reasoning and Thinking Traces:

  • Streams full model reasoning steps live in the side panel before executing browser actions.
  • Clean collapsible thought blocks provide transparency into what the agent is planning next.

---

Multiple windows and parallel tasks

Open a separate Chrome window and open OpenBUA there to give it a separate task. Each instance uses its own chat and browser window. Changing window focus does not redirect another instance's actions. Chats already open in another window cannot be selected or deleted until that window closes or switches chats.

The windows icon immediately before Send enables parallel subagents for the current chat. It is off by default; hover over it for a description. When enabled, OpenBUA can split independent work across up to three additional Chrome windows, using the selected model and API key. Workers inherit the task's approved documents and restrictions, keep separate histories, and return results to the parent. Sequential work stays in the parent window.

Each worker window includes an OpenBUA task tab showing its assignment, progress, result, and Stop button. Stop in the parent cancels its workers; closing a worker window or task tab cancels that worker. Browser windows remain available for reviewing their pages. All windows in the same Chrome profile share the profile's signed-in accounts.

For developers, npm run test:windows checks coordination and ownership. After building, npm run test:windows-extension tests the real extension with local forms and a simulated model; it requires Playwright with an extension-capable Chromium executable. Set OPENBUA_PLAYWRIGHT_MODULE and OPENBUA_EXTENSION_CHROME_PATH to override their locations.

AI Models and Supported Providers

OpenBUA supports both zero-cost free operation and flexible Bring Your Own Key (BYOK) providers:

Google Gemini Free Mode

• Zero-Cost Access: Try OpenBUA completely free without paid API subscriptions or credit card billing.
• Powered by Gemini 3.5 Flash Lite: Default high-limit free model with automatic rate-limit backoff and pacing (up to 1,500 requests per day).
• Direct Free Key Access: Built-in 1-click link to generate your personal free API key directly from Google AI Studio.
• Full Reasoning Support: Streams structured chain-of-thought blocks before each browser action.

Bring Your Own Key (BYOK)

• OpenAI Compatible: Connect to OpenAI (GPT-6 Astra, 6.1 Sol, 6 Luna, GPT-4o), Groq, OpenRouter, DeepSeek, or local Ollama instances.
• Anthropic Compatible: Connect to Anthropic (Claude Opus 5.5, Sonnet 5.5, Fable 5.1) or MiniMax.
• Role Normalization and Connection Testing: Built-in 1-click connection verification and automatic schema adaptation across providers.

---

Key Architectural Highlights

• Runs in Your Existing Browser: No headless emulators, remote browsers, or virtual containers. OpenBUA leverages your everyday authenticated sessions (Google, GitHub, LinkedIn, Twitter/X, internal intranets).
• 100% Client-Side and Zero Backend: Executes completely client-side in Chrome's Side Panel. Your API keys, browsing data, and documents never touch a third-party application server.
• Universal Browser Compatibility: Runs seamlessly on Google Chrome, Brave, Arc Browser, and Microsoft Edge with side panel support.
• Multi-Tab Chat Sessions: Create multiple concurrent chat sessions with individual persistent history stored in chrome.storage.local. Reloading or closing pages in Chrome never wipes your conversation history.
• Clean Dark Zinc Interface: Designed with Tailwind CSS shades of zinc, crisp Inter typography, and Apple-inspired accent elements.

---

Quick Installation Guide (For New Users)

No technical knowledge or coding required. Follow these steps to install OpenBUA in your browser in under 2 minutes:

Step 1: Download the Pre-Built Extension

1. Go to the Releases page.
2. Under the latest release, click on OpenBUA-extension.zip to download it.
3. Unzip the downloaded file to a folder on your computer. You will see a folder containing manifest.json, sidepanel.html, etc.

Step 2: Open Chrome Extensions and Enable Developer Mode

1. Open Google Chrome.
2. In the address bar, navigate to chrome://extensions and press Enter (or open the menu at top-right -> Extensions -> Manage Extensions).
3. In the top-right corner of the Extensions page, switch the Developer mode toggle to ON.
   • Note: "Developer mode" is a free built-in setting available in every copy of Chrome that allows loading local extensions.

Step 3: Load the Extension

1. After enabling Developer mode, click the Load unpacked button in the top-left corner.
2. Select the unzipped folder containing manifest.json (the folder extracted in Step 1).
3. Click Select Folder.
4. OpenBUA is now installed. Click the puzzle icon (Extensions) in your Chrome toolbar and pin OpenBUA for quick access.

Step 4: Configure Your AI Key

1. Click the OpenBUA icon on your Chrome toolbar to open the Side Panel.
2. Click the Settings tab.
3. Choose your configuration:
   • Google Gemini (Free): Click "Get a free Gemini API key", paste your key, and select Gemini 3.5 Flash Lite.
   • BYOK: Select OpenAI Compatible or Anthropic Compatible, paste your API key, and specify your model ID.
4. Click Test Connection to confirm connectivity, then click Save Settings.

---

Alternative: Build from Source (For Developers)

If you prefer to build the extension from source:

git clone https://github.com/AliSharjeell/OpenBUA.git
cd OpenBUA
npm install
npm run build

Then in chrome://extensions with Developer mode enabled, click Load unpacked and select the dist folder.

---

Architecture

OpenBUA/
├── manifest.json              # Chrome Manifest V3 configuration
├── sidepanel.html             # Side Panel HTML entry
├── vite.config.ts             # Vite build configuration
├── tailwind.config.js         # Dark-zinc palette and Inter font theme
├── scripts/
│   ├── build-extension.js     # Production build and bundling pipeline
│   └── generate-icons.js      # Extension icons generator
├── src/
│   ├── background/
│   │   └── service-worker.ts  # MV3 service worker and side panel launcher
│   ├── content/
│   │   └── content-script.ts  # DOM scanner, native input setter and event dispatcher
│   ├── agent/
│   │   ├── form-agent.ts      # Main agent harness wrapping pi-agent-core Agent
│   │   ├── stream-adapter.ts  # BYOK streaming adapter (OpenAI, Gemini, Anthropic SSE)
│   │   ├── tools.ts           # Browser use tool schemas and executors
│   │   └── browser-bridge.ts  # Chrome tabs and scripting communication bridge
│   ├── services/
│   │   ├── storage.ts         # Scoped local storage service (chrome.storage.local)
│   │   └── pdf-parser.ts      # In-browser PDF and text extractor
│   ├── components/
│   │   ├── ChatView.tsx       # Main chat interface with tool visualizer and file uploader
│   │   ├── MemoryView.tsx     # Global and Tab memory manager with toggle controls
│   │   ├── PreviewView.tsx    # Live research preview tab with streaming markdown
│   │   ├── SuggestedMemoriesView.tsx # Self-learning proactive memory suggestion view
│   │   ├── VaultView.tsx      # Secure credential and profile vault interface
│   │   ├── SettingsView.tsx   # Free Gemini and BYOK provider settings
│   │   ├── MarkdownRenderer.tsx # GitHub Flavored Markdown renderer with table repair
│   │   └── ui/                # UI component primitives (Button, Input, Card, Badge)
│   ├── sidepanel/
│   │   ├── App.tsx            # Application root with persistent chat session tabs
│   │   └── main.tsx           # React entry point
│   └── types/
│       └── index.ts           # TypeScript interfaces and contracts
└── test-form.html             # Multi-step test form for verification

---

Usage Examples

1. Autonomous Form Filling

Open any web form (or open test-form.html in Chrome) and click the Fill active form automatically button above the input bar, or ask:

Inspect the form on this page, match it against my stored profile, and fill in all fields.

OpenBUA will inspect the DOM, retrieve matching data from your memory store, fill the fields using native synthetic events, and report what was filled.

2. Deep Web Research with Live Preview

Navigate to any directory, search results page, or social platform and ask:

Search for 'world models PhD' researchers, go through the results pages, and extract a table containing name, current affiliation, and location.

Switch to the Preview tab in the top navigation to watch candidate rows stream into a formatted markdown table in real time as the agent navigates.

3. Real Email Outreach from Your Signed-in Email Client (Gmail, Outlook, Webmail)

Because OpenBUA operates right inside your existing browser, it can send genuine emails through your active, signed-in Gmail or Outlook account without asking for third-party OAuth permissions, SMTP passwords, or API configurations.

Simply open or let OpenBUA navigate to Gmail and instruct it:

Email Yann LeCun (yann.lecun@nyu.edu) to supervise our final year project. We are three FAST NUCES computer science students from Karachi, Pakistan building small world models with multi-token prediction. Draft a respectful email introducing our project and send it from my active Gmail account.

OpenBUA will:

• Switch to or open your active Gmail tab.
• Click Compose to trigger the Gmail draft window.
• Enter the recipient address and dispatch confirmed chip creation.
• Populate a clear subject line and draft a well-structured email body.
• Click Send to dispatch the email directly from your personal or professional mailbox.

---

Search & Discovery Keywords

Autonomous AI browser agent, browser-use Chrome extension, in-browser AI automation, automated web research, AI form filling, Claude Code for browser, local browser automation, Manifest V3 AI agent, web scraping assistant, self-learning agent memory, live research preview, BYOK browser agent.

---

Development

• Build: npm run build
• Typecheck: npm run typecheck
• Media transport tests: npm run test:chunks
• Development Server: npm run dev

---

Privacy Policy

OpenBUA is completely client-side. We do not operate external application servers and do not collect, track, log, or sell your browsing history, personal data, or form inputs. Read our full policy in PRIVACY.md.

---

License

MIT