Knowledge Networking Protocol

Find the friend you didn't know existed

No photos. No bios. No swiping. Match people by how they think — extracted from their markdown files and AI conversations.

Get Early Access Star on GitHub
scroll to explore
Demo Profile

How It Works

Three steps.
Zero friction.

Install a skill. See your mind. Find minds like yours.

1

Install the skill

Add /friends to your Claude Code session. One command. No signup, no app store, no credit card.

2

It reads your files locally

The skill scans your markdown files on your device. Extracts topic tags. Encodes them into a Bloom filter. Your files never leave your machine.

3

See your graph

An interactive map of minds that resonate with yours. Click a node. See their topics. Connect via Telegram. Start talking.

Privacy Architecture

Your files stay
on your device.

Topic extraction and encoding happen entirely on your device. The matching server receives only a Bloom filter — a fixed-size bit array that cannot be reversed to recover your original topics.


The server does not see, store, or process your files, text, or raw data. It computes similarity scores on compact bit arrays and returns your matches. That's it.


Your identity is secured by an Ed25519 keypair generated locally. Only you can update your profile. The private key never leaves your machine.

YOUR DEVICE | SERVER -------------------------------------+------------------ | Markdown files ─┐ | Claude sessions ─┤─ TF-IDF | │ extraction | ↓ | Topic tags | ↓ | Bloom filter ─────────|──→ Jaccard similarity (bit array) | on bit arrays | ↓ ←──────────────────|─── Top-K matches ↓ | Graph in browser | | Private key: ~/.friends/ | Stores: Bloom filter (never transmitted) | + public key + display name | Never stores: raw text, | files, topics, embeddings

Early Access

See your mind.
Find minds like yours.

We're building the first knowledge networking protocol. Be among the first to try it.

Star the repo on GitHub