AI-powered LinkedIn feed scanner that classifies your connections against custom goals.
LinkedIn Lens is a Chrome extension that auto-scrolls your LinkedIn feed, uses an LLM to extract and classify posts in real-time, and surfaces the people you should actually be talking to — with personalized talking points.
- You define "lenses" — free-text goals like "Find investors for my AI startup" or "Find donors for my friend's AI-for-good nonprofit"
- LinkedIn Lens scrolls your feed with randomized, human-like timing
- Each post's raw HTML is sent to an LLM (via OpenRouter) which extracts author info AND classifies them against your lenses in a single call
- Results appear in a live sidebar — ranked by confidence, with talking points for each match
Most LinkedIn scrapers break every time LinkedIn changes their CSS classes. LinkedIn Lens takes a different approach: it sends sanitized HTML directly to the LLM and lets the model figure out the structure. No hardcoded selectors. No regex. The LLM understands HTML natively.
LinkedIn DOM → strip noise (SVGs, buttons, styles) → send HTML to LLM → structured data + classification
One API call does both extraction and classification. Resilient by design.
| Sidebar with matches | Sidebar detail | Settings |
|---|---|---|
|
 |
 |
┌─────────────────┐ ┌──────────────────┐ ┌─────────────┐
│ Content Script │────▶│ Service Worker │────▶│ OpenRouter │
│ (LinkedIn tab) │ │ (Background) │ │ (LLM API) │
│ │ │ │ │ │
│ • Find HTML │ │ • Queue mgmt │ │ • Extract │
│ chunks │ │ • Rate limiting │ │ • Classify │
│ • Auto-scroll │ │ • Person merge │ │ • Score │
│ • Dedup │ │ • Storage │ │ │
└────────┬─────────┘ └──────────────────┘ └─────────────┘
│
▼
┌─────────────────┐
│ Shadow DOM │
│ Sidebar │
│ │
│ • Ranked cards │
│ • Talking points│
│ • Lens filters │
│ • Export │
└─────────────────┘
- Manifest V3 — service workers, not background pages
- Persistent queue — survives service worker restarts with exponential backoff
- Person aggregation — merges sightings across sessions by profile URL
- Shadow DOM sidebar — CSS-isolated from LinkedIn's styles
The extension ships with three example lenses (fully editable):
| Lens | Description |
|---|---|
| Nonprofit Supporters | Donors and allies for an AI-for-good nonprofit |
| Startup Investors | Angels and VCs for an AI startup |
| Design Partners | Event organizers and community leads |
- Clone this repo
- Open
chrome://extensionsin Chrome - Enable Developer mode (top right toggle)
- Click Load unpacked and select the repo folder
- Go to the extension's Options page
- Enter your OpenRouter API key
- Edit your lenses (goals)
- Navigate to
linkedin.com/feed— the sidebar will appear
- API Key: Your OpenRouter API key (stored locally)
- Model: Default is
google/gemini-3.1-flash-lite-preview— fast and cheap. Any OpenRouter model works. - Lenses: Free-text goal descriptions. The LLM interprets them — be as specific as you want.
- Auto-scroll: 5-15 second randomized delays, 2-minute session cap, stops on manual scroll
This extension was built in a single live session using:
- Claude Code — Anthropic's AI coding CLI
- oh-my-claudecode — Multi-agent orchestration layer
- Pipeline: Deep Interview → Ralplan (consensus planning) → Autopilot (parallel execution)
- Zero lines of code typed by hand
| Metric | Value |
|---|---|
| Agents spawned | 12 (7 types) |
| LLM API calls | 154 |
| Total tokens | 13.4M (97% cache reads) |
| Actual output generated | 111K tokens |
| Tool calls | 163 |
| Total cost | ~$34 |
| Build time | ~2 hours |
| Lines typed by hand | 0 |
Agents used: Planner ×2, Architect ×2, Critic ×2, QA Tester ×3, Explorer ×1, Code Reviewer ×1, Scientist ×1
Cost breakdown: $19 cache reads · $8 output generation · $6 cache writes
Read the full workshop report →
- All data stays in Chrome's local storage
- Your API key is stored locally, never sent anywhere except OpenRouter
- No tracking, no analytics, no external servers
- LinkedIn HTML is stripped of noise before being sent to the LLM
MIT