Taar · تار

Weave your network. Smart, human-like LinkedIn networking automation — built on Bun & Electrobun.

Features · Quick Start · How It Works · FAQ · Contributing

---

Taar (Persian: the warp threads of a carpet) is a lightweight desktop app that opens an isolated browser, sends connection requests to people-search results on your behalf, and behaves like a careful human while doing it — randomized delays, per-session behavior personas, exponential cooldown when something goes wrong, and a selector engine that learns what works on LinkedIn's changing DOM.

Built on Bun and Electrobun instead of Electron: smaller footprint, faster startup, one runtime end to end.

✨ Features

🖥 Tiny desktop shell	Electrobun 2.0 + Bun — no Electron, no bundled Chromium; uses your installed Chrome/Edge
🧠 Adaptive pacing	Exponential cooldown after consecutive errors; successes reset it. Protects your account
🎭 Behavior personas	Every session picks a rhythm at random (steady / brisk / careful) — no two runs look alike
🔍 Self-learning selectors	Ranks and remembers which DOM strategies find Connect buttons; survives LinkedIn redesigns
⏭ Pending-aware	Detects and skips "Pending" (withdraw) invitations instead of timing out on them
🛡 Limit detection	Stops automatically on weekly-limit or restriction screens (English + Persian)
💾 Persistent sessions	Log in once — cookies and storage are saved and reused
🔄 Crash-proof relaunch	Browser liveness watchdog + force-restart + self-healing UI: closing the browser never dead-ends the app
📝 Personalized notes	Optional note per invitation with quota tracking and smart truncation
🪞 Persian-inspired UI	Boteh-jeghe patterns, gold tashir accents, carpet-border cards — with an in-app guide

🚀 Quick Start

Prerequisites

• Bun 1.1+
• Google Chrome (or Edge) installed on the system
• A LinkedIn account

Install & Run

git clone https://github.com/MHAlikhani/LinkedIn-Assistant.git
cd LinkedIn-Assistant
bun install
bun start

The first run downloads the Electrobun runtime once. Then:

1. Click Launch Browser and log in to LinkedIn in the window that opens.
2. Navigate to linkedin.com/search/results/people.
3. Set the number of requests (20–30 is a safe daily range for free accounts).
4. Click Start.

Build for distribution

bun run build

📦 How It Works

┌─────────────────────────────────────────────┐
│  Electrobun main process (Bun runtime)      │
│  ├── typed RPC ⇄ webview UI                 │
│  └── spawns ──► Bot child process (Bun)     │
│                  ├── DecisionEngine         │
│                  ├── HumanSimulator         │
│                  ├── SmartElementFinder     │
│                  ├── ModalStateMachine      │
│                  ├── NavigationGuard        │
│                  └── Playwright (Chrome)    │
└─────────────────────────────────────────────┘

The bot runs in an isolated child process and talks to the UI over a typed RPC bridge. When you close the browser, the watchdog notices within seconds, the process exits cleanly, and the Launch button comes back — no app restart needed.

⚙️ Configuration

Environment variables:

Variable	Default	Description
BOT_RUNTIME	bun	Set to node to run the bot child process on Node.js ≥ 20
LOG_LEVEL	info	debug · info · warn · error

Data files (created on demand in the project folder):

File	Purpose
data/session.json	Saved LinkedIn login session
data/selector-memory.json	Learned selector rankings

🧪 Development

bun test            # 81 tests across 9 suites
bun run lint        # ESLint
bun start           # dev launch (hot rebuild)
bun run clean       # remove reports + learned data

Project layout:

src/
├── bun/          Electrobun main process (window, RPC handlers)
├── runtime/      Bot process manager + runtime resolver
├── schemas/      Shared RPC schema
├── mainview/     Webview UI (HTML/CSS/JS, Electrobun RPC client)
└── bot/          The automation brain
    ├── core/     LinkedInBot, DecisionEngine, HumanSimulator,
    │             ModalStateMachine, NavigationGuard, SessionManager
    ├── selectors/  Adaptive selector engine + learned memory
    └── utils/    IPC logger, sleep helpers
docs/             Architecture notes and ADRs
tests/            Unit + integration tests (bun test)

❓ FAQ

Is this against LinkedIn's rules? Automated connection requests violate LinkedIn's User Agreement. Use it responsibly on your own account, keep volumes low, and understand the risk of account restriction. This software is provided for educational purposes.

Why does it move the mouse and pause randomly? Delays are Gaussian-distributed, mouse paths follow eased curves, and each session adopts a different pacing persona — the goal is to behave like a tired human, not a for loop.

The bot stopped after a few invitations. Check the log: a weekly-limit or restriction screen was detected and the bot stopped to protect your account.

I closed the browser window. Just click Launch Browser again — recovery is automatic.

Does it work on macOS/Linux? The stack is cross-platform (Electrobun supports all three). The bot resolves Chrome/Edge the same way; please report issues you hit on your OS.

📚 Documentation

• Architecture
• ADRs — decision records, including the Electrobun migration
• Changelog

🤝 Contributing

Issues and PRs are welcome. Keep PRs focused; run bun test and bun run lint before submitting.

📄 License

Taar Community License 1.0 — © Mohammad Hossein Alikhani

You can…	For free?
Use Taar for your own personal networking	✅ Yes, unlimited
Study and modify the code for personal use	✅ Yes
Contribute pull requests to the official repo	✅ Yes (and welcome!)
Share verbatim copies pointing to this repo	✅ Yes
Commercial / industrial / employer use	💰 Needs a paid license
Re-publish under a different name or rebrand	💰 Needs a paid license
Embed Taar's code in another product	💰 Needs a paid license

Commercial licensing: mohammad.hosein.alikhani08@gmail.com — see LICENSE.

---

_{LinkedIn automation · LinkedIn bot · connection requests · networking automation · Playwright · Bun · Electrobun · desktop app · human-like automation · adaptive selectors}