Robin Dunbar figured the human brain caps out at roughly 150 stable relationships. After that, people just fall off. Not because you stop caring, but because you can't track that many threads at once. I built PingCRM to push past that limit, not by making you remember more, but by handling the part you're bad at: noticing who's going cold and drafting something worth sending.

CRMs exist, but they're built for sales teams. I wanted something for one person maintaining their own network. Specifically:

(1) Pull in conversations I'm already having. Gmail threads, Telegram DMs, Twitter DMs, LinkedIn messages, no manual logging.

(2) Tell me who's going cold. A relationship score (0–10) based on recency, frequency, reciprocity, and breadth of interaction.

(3) Draft the message for me. Claude reads your conversation history, picks the right tone (formal vs. casual), and writes a follow-up that references why now is a good time, like a job change, a long silence, or a fundraising announcement they just posted.

How it works:

Connect your accounts in Settings. A Chrome extension handles LinkedIn (calls the Voyager API from your browser, cookies never leave the extension). Background sync runs on Celery beat. Gmail every 6h, Telegram and Twitter daily, LinkedIn on page visit.

The suggestion engine runs once a day: picks 5 contacts from two pools (active relationships going cold + dormant contacts worth reviving), applies cooldowns so you're not pinging someone every week, and generates AI drafts.

Identity resolution merges the same person across platforms. Deterministic on email/phone, probabilistic on name+company+username similarity.

Stack: Python/FastAPI, PostgreSQL, Redis/Celery, Next.js 15/React 19, Anthropic Claude API. ~1,100 tests across backend and frontend.

Self-hostable via Docker Compose (one command) or manual setup. AGPL-3.0. A hosted version is coming for people who don't want to run infrastructure.

Repo: https://github.com/sneg55/pingcrm

Docs: https://docs.pingcrm.xyz

Landing: https://pingcrm.xyz

I'd love feedback on the approach. The thing I'm least sure about is whether 5 suggestions/day is the right number, and whether the scoring formula actually captures what matters in a relationship. Happy to talk architecture or answer questions.