flockml 1.0.0 → 1.0.1

package/README.md

@@ -0,0 +1,72 @@
+# FlockML 🚀
+
+**Decentralized, WebGPU-Accelerated Machine Learning in the Browser.**
+
+FlockML allows you to train massive Machine Learning models (100M+ parameters) entirely in the background of your website visitors' browser tabs. It uses idle compute to create a free, decentralized supercomputer.
+
+[![npm version](https://img.shields.io/npm/v/flockml.svg)](https://www.npmjs.com/package/flockml)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
+
+## 🔥 Features
+- **WebGPU Native Acceleration:** Bypasses V8 JavaScript limits by executing matrix math directly on the user's GPU via TensorFlow.js.
+- **Zero-Latency Payloads:** Uses Protocol Buffers and 8-bit quantization to compress network layer weights down to 91 bytes.
+- **Differential Privacy:** Mathematically secures user data by injecting cryptographic Laplacian noise into gradients before they leave the browser.
+- **Server-Side Anomaly Detection:** Instantly drops poisoned payloads (`NaN` injection, Sybil attacks) to protect the global model.
+
+## 📦 Installation
+
+```bash
+npm i flockml
+```
+
+## 🚀 Quick Start
+
+FlockML uses a **Federated Learning** architecture. You need a centralized Coordinator (Server) and Edge Nodes (Clients).
+
+### 1. Initialize the Server Coordinator (Node.js)
+The coordinator manages the global weights and runs the `FedAvg` aggregation algorithm.
+
+```typescript
+import { Coordinator } from 'flockml';
+
+// Initialize a network with 2 inputs, 4 hidden nodes, 1 output
+const coordinator = new Coordinator(2, 4, 1);
+
+// Send these weights to your clients via WebSocket
+const globalWeights = coordinator.getGlobalWeightsForBroadcast();
+
+// When a client sends back an updated payload:
+// coordinator.receiveUpdate(binaryPayload);
+// coordinator.aggregate();
+```
+
+### 2. Run the Edge Node (Browser)
+The edge node runs completely sandboxed in a Web Worker, utilizing WebGPU to train the model locally.
+
+```typescript
+import { FlockNode } from 'flockml';
+
+// 1. Initialize the edge node
+const node = new FlockNode(2, 4, 1);
+
+// 2. Sync with the server's global weights
+node.syncGlobalWeights(weights_ih, weights_ho, bias_h, bias_o);
+
+// 3. Train on the user's local data
+const inputs = [[0, 0], [0, 1], [1, 0], [1, 1]];
+const targets = [[0], [1], [1], [0]];
+node.trainLocalBatch(inputs, targets);
+
+// 4. Export secure, quantized protobuf payload to send back to server
+node.privacyEpsilon = 1.0; // Enable Differential Privacy
+const binaryPayload = node.exportSecureGradients(); 
+```
+
+## 📚 Documentation
+For deep profiling reports, stress tests, and advanced architecture guides, visit the [official FlockML documentation](https://supratimdev.qzz.io/flock-ml).
+
+## 🛡️ Security
+FlockML strictly executes inside isolated Web Workers, ensuring it has zero access to the DOM, cookies, `localStorage`, or the client’s file system.
+
+## 📄 License
+MIT License

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "flockml",
-  "version": "1.0.0",
+  "version": "1.0.1",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
   "scripts": {