📋 Project Overview

What It Does

LinkedIn Automator is a personal cold outreach tool that runs entirely on your local machine. It reuses your existing Chrome login session to operate LinkedIn as you would — no headless scraping, no unofficial APIs.

It handles three phases of the outreach funnel automatically:

1. Find decision makers at target companies
        ↓
2. Send AI-personalized connection request notes
        ↓
3. Send AI-personalized first messages to new connections
        ↓
4. Detect replies — send AI-personalized follow-up if silent

Tech Stack

Layer	Technology	Role
Backend	Python + FastAPI	API server, WebSocket streaming
Browser	Playwright (Chromium)	LinkedIn automation via persistent Chrome context
AI	Ollama (`llama3.1:8b`) or Gemini	Message generation
Frontend	Vanilla JS + CSS	Single-page dashboard UI
Data	CSV files	Logging, dedup, follow-up tracking
Realtime	WebSockets	Live log streaming to UI

File Map

linkedin-automator/
├── main.py           # FastAPI app — all HTTP + WebSocket endpoints
├── automator.py      # Connection request automation (core Playwright logic)
├── messenger.py      # Messaging + follow-up automation
├── logger.py         # All CSV read/write (connections, messages, followups)
├── ai.py             # Ollama + Gemini wrappers
├── run_logger.py     # Per-run history logging
├── .env              # Config (AI provider, caps, Chrome path)
│
├── static/
│   ├── index.html    # Dashboard UI
│   ├── app.js        # All frontend JS
│   └── style.css     # Styling
│
├── connections.csv   # Log of all connection requests sent
├── messages.csv      # Log of all first messages sent
├── followups.csv     # Follow-up tracker with status per person
│
├── messenger_debug.log  # Verbose step-by-step automation log
└── obsidian-vault/      # This documentation

Key Design Decisions

IMPORTANT — Persistent Chrome Context The tool uses launch_persistent_context() which reuses your real Chrome profile. LinkedIn sees it as your normal browser session — not automation. This is the primary reason it works without triggering bot detection.

NOTE — No Headless Mode Chrome runs visibly. You can watch every action in real time and stop anything with the Stop button.

TIP — Local AI Only (default) By default, Ollama runs models locally. Your outreach data and message content never leave your machine.