Implements a privacy-preserving financial learning system that runs entirely
in the browser using WebAssembly. Key features:
- PlaidLocalLearner: Browser-local ML engine with IndexedDB persistence
- Q-learning for budget optimization and spending recommendations
- HNSW vector index for semantic transaction categorization
- Spiking neural network for temporal pattern recognition
- Anomaly detection for unusual transaction flagging
- Zero data exfiltration - all learning stays client-side
Components:
- examples/edge/src/plaid/mod.rs: Core Rust learning algorithms
- examples/edge/src/plaid/wasm.rs: WASM bindings for browser
- examples/edge/pkg/plaid-local-learner.ts: TypeScript API wrapper
- examples/edge/pkg/plaid-demo.html: Interactive demo page
- examples/edge/docs/plaid-local-learning.md: Comprehensive documentation
Privacy guarantees:
- Financial data never leaves the browser
- Optional AES-256-GCM encryption for IndexedDB storage
- User can delete all data instantly
- No analytics, telemetry, or tracking
- Build dual WASM targets (web + nodejs) for universal compatibility
- Add Node.js polyfills for web APIs (crypto, performance, window, document)
- Create universal entry point with auto-detection of environment
- Update CLI with comprehensive benchmark, demo, and info commands
- Fix ESM/CJS compatibility with .cjs extension for Node.js module
- Package includes both browser and Node.js WASM binaries
Published to npm as @ruvector/edge-net v0.1.1
Package: 885.4 kB compressed, 3.2 MB unpacked
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Implement comprehensive tests for adaptive learning mechanisms including MicroLoRA and SONA in learning_tests.rs.
- Introduce tests for bio-inspired neural components such as HDC, BTSP, and Spiking Neural Networks in nervous_system_tests.rs.
- Create common utilities for random vector generation, vector assertions, and softmax calculations in mod.rs.
- Ensure all tests validate expected behaviors and maintain numerical stability.
Four attention mechanisms answering fundamental questions:
- Neural Attention: What words/tokens matter?
- DAG Attention: What computational steps matter?
- Graph Attention: What relationships matter?
- State Space: What history still matters?
Includes:
- dag_attention.rs: Critical path analysis, topological ordering
- attention_unified.rs: Unified interface composing all 4 types
- Updated mod.rs architecture diagram
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Enable capabilities module with pub export
- Add compute/ module with SIMD, WebGPU, WebGL backends
- Add ai/ module with attention, router, federated learning, LoRA
- Streamline WASM API for Time Crystal, NAO, MicroLoRA, HDC, WTA, BTSP
- Add Global Workspace and Morphogenetic network support
- Add learning scenarios for error recovery and file sequences
- Add swarm collective intelligence and consensus modules
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Critical security fixes before production deployment:
1. Argon2id PBKDF in pikey/mod.rs (replaces SHA-256)
- Memory-hard KDF with 64MB memory, 3 iterations
- Version 0x02 format with salt, backward compatible with v1
- Secure zeroization of key material
2. Ed25519 signature verification in rac/mod.rs
- Real cryptographic verification for authority resolutions
- ScopedAuthority::sign_resolution() helper for signing
- Canonical message format for verification
3. Password-protected key export in identity/mod.rs
- export_secret_key now requires 8+ character password
- AES-256-GCM encryption with Argon2id-derived key
- import_secret_key for secure recovery
Dependencies added:
- argon2 v0.5 (memory-hard KDF)
- zeroize v1.7 (secure memory cleanup)
Test coverage:
- 125 tests passing (40 lib + 85 integration)
- Updated adversarial tests with real Ed25519 signatures
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
## Learning Module (src/learning/mod.rs)
- ReasoningBank for pattern storage with similarity lookup and pruning
- TrajectoryTracker ring buffer for task execution tracking
- Spike-driven attention for 87x energy efficiency (based on Yao et al.)
- Multi-head attention for distributed task routing
- NetworkLearning unified interface for edge nodes
## RAC Module (src/rac/mod.rs) - Adversarial Coherence Thesis
Implements the 12 axioms for browser-scale adversarial truth maintenance:
1. Connectivity is not truth
2. Everything is an event
3. No destructive edits (deprecation only)
4. Every claim is scoped
5. Semantics drift is expected
6. Disagreement is signal
7. Authority is scoped, not global
8. Witnesses matter
9. Quarantine is mandatory
10. All decisions are replayable
11. Equivocation is detectable
12. Local learning is allowed
Core components:
- Append-only Merkle event log for tamper-evident history
- CoherenceEngine for conflict detection and resolution
- QuarantineManager for contested claims
- Authority policy and verifier traits
- Decision traces for audit and replay
## Integration
- Learning and RAC integrated into EdgeNetNode
- 28 tests pass (13 new tests for learning/RAC)
References:
- FLP Impossibility (MIT CSAIL)
- PBFT Byzantine Fault Tolerance
- CRDTs (Lip6)
- RFC 6962 Certificate Transparency
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Updated README to focus on research and simulation aspects:
- Renamed to 'Artificial Life Simulation'
- Cells instead of nodes terminology
- Energy system instead of cryptocurrency
- Clear disclaimer: NOT a financial product
- Research goals and applications
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add Concurrency section with 4 worker modes:
- Main Thread (no workers, simple)
- Worker Pool (auto-scaling, recommended)
- Dedicated (one worker per task type)
- Shared Worker (cross-tab coordination)
- Add comprehensive worker templates with code examples
- Update stats to show worker count
- Include WorkerPool class with batch operations
- Add SharedWorker cross-tab example
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add comprehensive Web Workers section explaining UI responsiveness
- Include WorkerPool usage examples with practical code
- Add feature table explaining auto-scaling, load balancing, timeouts
- Add "when to use workers" guidance table
- Update Table of Contents with Consensus Modes and Web Workers
- Add WorkerPool to API Reference section
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add npm package section at top with install commands
- Link to pkg/README.md for JavaScript documentation
- Clarify Raft vs Gossip+CRDT consensus modes
- Add Web Worker pool to distributed systems features
- Update WASM badge to show 364KB size
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Include worker.js and worker-pool.js in package
- Add exports for @ruvector/edge/worker and @ruvector/edge/worker-pool
- Document WorkerPool API with examples
- Features: round-robin distribution, batch splitting, load balancing
- Keeps UI responsive during heavy WASM operations
- Bump to v0.1.9
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Raft assumes stable membership and trusted nodes - not suitable for
wild browser swarms. Updated docs to:
- Position Raft for "trusted cohorts" (teams, enterprise, private relays)
- Add Gossip + CRDT for "open swarms" (public, high-churn, adversarial)
- Explain when to use each mode with code examples
- Update capability tables to reflect both consensus strategies
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Update heading to "Free Self-Learning AI Swarms at the Edge"
- Emphasize self-optimizing agents that get smarter over time
- Mention LoRA fine-tuning, EWC++ continual learning, ReasoningBank
- Bump to v0.1.7
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add comprehensive platform diagram showing edge vs edge-full
- List all capabilities of edge-full modules
- Add optional/peer dependency on @ruvector/edge-full
- Show usage patterns for both packages together
- Bump to v0.1.6
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- 5-step tutorial: Identity, Crypto, Vector Search, Routing, Consensus
- P2P transport options: WebRTC, GUN.js, IPFS/libp2p, Nostr
- Full code examples for each transport integration
- Architecture diagrams showing edge-first design
- Transport comparison table with recommendations
- Complete API reference for all 9 WASM exports
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Emphasize zero-cost edge-first architecture
- Add economics comparison (cloud vs edge)
- Document all WASM APIs with examples
- Add use cases and architecture diagram
- Compare with cloud alternatives (OpenAI, LangChain, AutoGPT)
- Include agentic-flow integration example
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Built WASM bindings with wasm-pack for browser/Node.js usage
- Exports: WasmIdentity, WasmCrypto, WasmHnswIndex, WasmSemanticMatcher,
WasmRaftNode, WasmHybridKeyPair, WasmSpikingNetwork, WasmQuantizer,
WasmAdaptiveCompressor
- 364KB optimized WASM binary with full TypeScript types
- Published to npm as @ruvector/edge@0.1.0
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add CLI tool with run, backtest, paper trading, analyze, and benchmark
- Add visualization module with ASCII charts (line, bar, sparkline, table)
- Create Jest test suite covering all production modules
- Implement example strategies: Hybrid Momentum, Mean Reversion, Sentiment
Performance benchmarks show all modules production-ready:
- Kelly Engine: 0.014ms (71,294/s)
- LSTM-Transformer: 0.681ms (1,468/s)
- DRL Portfolio: 0.059ms (17,043/s)
- Sentiment Alpha: 0.266ms (3,764/s)
Backtesting:
- Single-pass metrics calculation (was 10+ passes)
- Inline stats: mean, variance, win/loss counts computed together
- Combined drawdown metrics in one pass
- Removed redundant method calls
Risk Management:
- Ring buffers for trade history (O(1) vs O(n) shift/slice)
- Running sum for volatility average (O(1) vs O(n) reduce)
- Incremental loss count tracking
Reduces iteration overhead by ~5-10x for large datasets.
- Added SIMD badge and documentation
- Added ParallelEmbedder API reference and usage examples
- Updated performance benchmarks with parallel vs sequential comparison
- Added browser compatibility table
- Added changelog section
- Added batch processing use case example
- Updated build instructions with SIMD flags
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- ParallelEmbedder class using Node.js worker_threads
- Distributes batches across multiple CPU cores
- Benchmark results: 3.6-3.8x speedup on batch processing
- Per-text latency drops from ~390ms to ~103ms with 4 workers
- Published v0.1.2 to npm and crates.io
Usage:
import { ParallelEmbedder } from 'ruvector-onnx-embeddings-wasm/parallel';
const embedder = new ParallelEmbedder({ numWorkers: 4 });
await embedder.init();
const embeddings = await embedder.embedBatch(texts);
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Enable WASM SIMD128 instructions for vectorized operations
- Update simd_available() to properly detect SIMD at compile time
- SIMD build is 180KB smaller than non-SIMD (more compact instructions)
- Published v0.1.1 to both npm and crates.io
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Added npm and crates.io version badges
- WebAssembly and MIT license badges
- Quick start examples for Browser, Node.js, and Cloudflare Workers
- Complete API reference for WasmEmbedder, WasmEmbedderConfig
- Model comparison table with 6 HuggingFace models
- Performance benchmarks and use case examples
Published to npm as ruvector-onnx-embeddings-wasm@0.1.0
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- gnn-correlation-network.js: Added RollingStats class for O(1) incremental
updates and correlation caching with TTL to avoid redundant O(n²) calculations
- attention-regime-detection.js: Optimized matmul with cache-friendly i-k-j
loop order and added empty matrix guards
- quantum-portfolio-optimization.js: Added ComplexPool for object reuse to
reduce GC pressure, plus in-place operations (addInPlace, multiplyInPlace,
scaleInPlace) to avoid allocations in hot loops
- multi-agent-swarm.js: Added RingBuffer for O(1) bounded memory operations
and SignalPool for signal object reuse
New optional companion package using Tract for inference:
- Runs in browsers, Cloudflare Workers, Deno, edge environments
- Same API as native crate
- JavaScript bindings via wasm-bindgen
- Supports all pooling strategies (Mean, Cls, Max, etc.)
Uses Tract instead of ONNX Runtime for WASM compatibility.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Changed invalid category "machine-learning" to "algorithms".
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
