🔧 Configuration & Setup

← Home

---

Prerequisites

• Python 3.11+
• Google Chrome installed
• Ollama running locally (or Gemini API key)
• LinkedIn account (logged in via Chrome)

---

Installation

cd ~/LI/linkedin-automator
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
playwright install chromium

---

.env Reference

# AI Provider: "ollama" or "gemini"
AI_PROVIDER=ollama

# Ollama settings
OLLAMA_MODEL=llama3.1:8b
OLLAMA_BASE_URL=http://localhost:11434

# Gemini settings (if using Gemini)
GEMINI_MODEL=gemini-2.0-flash-exp
GEMINI_API_KEY=your_key_here

# LinkedIn caps
DEMO_CAP=3
TOTAL_HOURS=10

# Chrome profile (auto-detected if blank)
CHROME_PROFILE_PATH=/Users/jkw/Library/Application Support/Google/ChromeLinkedIn

# Fuzzy match threshold for company search (0-100)
FUZZY_MATCH_THRESHOLD=80

WARNING — Chrome Profile The Chrome profile at CHROME_PROFILE_PATH must be logged into LinkedIn. The tool will fail silently if Chrome opens to a login page.

---

Starting the Server

cd ~/LI/linkedin-automator
source venv/bin/activate
uvicorn main:app --reload --port 8000

Then open http://localhost:8000 in your browser.

---

AI Provider Setup

Option A — Ollama (local, recommended)

# Install Ollama
brew install ollama   # macOS

# Pull the model
ollama pull llama3.1:8b

# Start Ollama (auto-starts on macOS)
ollama serve

Test via the Test AI button in the dashboard.

Option B — Gemini

1. Get API key from Google AI Studio
2. Set in .env:

   AI_PROVIDER=gemini
   GEMINI_API_KEY=your_key_here
   

3. No local model needed

---

Chrome Profile Setup

The tool needs a dedicated Chrome profile that's logged into LinkedIn.

# Create a dedicated profile (run this once)
/Applications/Google\ Chrome.app/Contents/MacOS/Google\ Chrome \
  --user-data-dir="/Users/jkw/Library/Application Support/Google/ChromeLinkedIn" \
  --no-first-run

Log into LinkedIn in that Chrome window, then close it. The tool will reuse the session.

---

Speed Mode Recommendations

CAUTION — LinkedIn Rate Limits LinkedIn will restrict your account if you send too many requests too quickly. Use SAFE mode for production runs. Never exceed ~20 connections/day or ~5 messages/hour.

---

Data Files

All data is stored locally. Never committed to git (*.csv in .gitignore).

Manually adding a row to messages.csv

If you messaged someone outside the tool:

2026-02-20 16:00,Marissa,,,Message sent manually — logged to prevent duplicate outreach.

The dedup system will then skip them automatically.